### Summary

This PR adds **AI preview cards** for website records. When an AI agent returns supported website content, the chat can now show rich, clickable cards instead of only plain record names.

Supported records include products, product variants, events, blog posts, appointment types, courses, lessons, and jobs.

### What changed

#### Record preview flow

Add `_ai_tool_prepare_record_previews` to prepare AI search results for display before the final answer is posted.

The tool keeps the result order, stores preview links in the message body, and exposes metadata that the frontend can use to either render visual cards or keep normal records' links visible when cards are not available.

#### Website card rendering

Introduce `ai.preview.card.mixin` for website models that support visual previews.

Each website bridge module provides its own `_ai_get_preview_cards_render_context`, so preview cards reuse the same public website templates and styling as the corresponding website pages.

#### Frontend integration

Add the `AIPreviewCardSet` component, preview-card cache, styling, and carousel behavior.

Cards are rendered in website AI chat and embedded livechat. For embedded livechat, cards are injected through a light-DOM slot so website styles can still apply across the livechat shadow DOM.

#### Website-aware pricing

Add `ai.product.pricing.mixin` to expose website-aware pricing metadata for products, appointments, and paid slide channels.

---

task-id-5153868

When a payslip is corrected, the resulting correction payslip represents an adjustment, not a full re-statement of earnings. Showing absolute values on the PDF was confusing: the employee would see the same gross/net figures as the original, making it unclear what actually changed. 

This PR makes correction payslips display the difference from the original in the PDF report (worked days, salary lines, totals), so it is immediately clear what was adjusted and by how much. On a correction PDF report, the Qty, Amount, and Rate columns are handled:
- If only one of these three fields has changed, the report shows the delta for that field and the original values for the others, so that the product of the columns matches the total delta.
- If two or more fields has changed, only the Total delta is shown, as a calculated product of multiple deltas would be misleading.

To support this, two computed fields are introduced on hr.payslip: is_correction_payslip (True when the slip has an origin and is not a refund) and correction_net_delta (the net wage difference vs. the origin). 

A few related improvements are bundled:
- The reverted (refund) payslip is automatically validated and created as a refund of the origin, while the correction payslip is left in draft.
- Reverted payslips are excluded from PDF generation and e-mail notification.
- When paying a correction, the system groups the amount between the correction and its corresponding reverted payslip and marks both as paid once the payment is confirmed (unless the reverted payslip was already paid).
- When a negative correction delta occurs (the employee was overpaid), the workflow uses the existing negative-net warning: the correction sets 'has_negative_net_to_report' and 'negative_net_to_report', so the usual warning and salary attachment flow will handle recovery on next payslips.
- The "wrong version" and issues detection now accounts for worked-day-level versions and time-off changes on already-paid payslips.
- The related payslips smart button has been extended to corrections and reverts to access their origin and its related slips.

The slot selection page in appointment has been changed a large
number of times. Adding a lot of small features to a complex flow
made the code a mix of old and new, mixing interaction features
and DOM manipulations. Parts were starting to get outdated, and
the flow extremely complex to maintain, as comporting a lot of
variation depending on the appointment type setup.

Main changes
---

In order to modernize the UI of the page, as well as to provide a
cleaner mobile experience, we redesign it completely, rebuilding
the js code and templates around the existing fundations.

- The calendar is now foldable in a dropdown
- The user / resource selection is harmonized: cards are used for
both values of select_first (entity / date)
- Selection is also foldable in a dropdown when starting with the
entity.
- Loaders and helpers are harmonized and streamlined into fewer
cases / elements on the page.
- When capacity is managed, 2 is selected as default if possible. We
fallback to 1 otherwise.
- The two-column design is abandoned.

Also, some auto-confirmation is implemented:

- When clicking on an entity, when it is last thing to select, submit
the slot selection.
- When selecting the time for the slot, submit the selection if no
additional step is required.

Technical Improvements
---

- The previous way to refresh the availability was to rerender the full
calendar template in the controller, return it and replace its outerHtml
in the page. This is not dynamic nor interactive. We changed this old
behavior to a more classical approach: the controller only returns a
JSON with relevant data, and we render the calendar directly from
the JS. We therefore moved the template into a JS one.
- Controllers are cleaned from redundant data, and reworked to match
previous change.
- JS is streamlined as much as possible, reducing unjustified differences
between similar elements. Interaction features are exploited as much
as possible

Other changes
---
- Currently, we only compute one month of availability at a time for
performance reasons. Keep this behavior, but replace the context
key by real arguments in the controller. Also enable this behavior for
flexible appointments
- Remove 'discard' button when adding guests
- Some ui polishing of various views in the booking flow
- Added a small summary, only visible on mobile, at the top of the
registration page. This allows a quick look in the booking details
on mobile before starting filling details.

Task-5358867

Issue:
Cancelled journal entries were visible in the "Journal Items to Reconcile" view (action_move_line_posted_unreconciled) because the action domain had no filter to exclude them. This caused reconciliation failures, no reconciliation happened, and the cancelled record remained in the view.

This regression was introduced during a refactor to allow draft entries in the reconciliation view, where the posted-state condition was removed from the domain:

Enterprise commit: https://github.com/odoo/enterprise/commit/003cffabda7d91a6d10d58942ed972ca5e17366d

As a result, cancelled journal items also became visible, causing reconciliation attempts to fail while the records remained in the view.

Also, we are not allowed to reconcile cancelled move lines, and we already
have the validation for this [here](https://github.com/odoo/odoo/blame/a236f67776616f6facdefb0117a6ffdde9b7c84c/addons/account/models/account_move_line.py#L2627)

Issue is reproducible on runbot. Here is the video reference: https://drive.google.com/file/d/1ojIDxHn5Yst8gVFy8JyhwtJoDSSSJsmK/view?usp=sharing

- OPW: 6247870

Forward-Port-Of: odoo/enterprise#119017
Forward-Port-Of: odoo/enterprise#118773

Before this fix:
The credit limit warning calculation only considered credit notes but ignored outstanding bank payments 
when computing the partner's effective outstanding balance. For example, if a customer had a credit limit
of 1,000 and an invoice of 2,000 was created, then a bank payment of 1,500 was received, the warning 
would still incorrectly appear showing the customer exceeded their limit (2,000 > 1,000), even though the
actual outstanding amount was only 500.

After this fix:
The credit limit warning now properly includes outstanding bank payments in the calculation. Two cases 
are handled:
- Bank payments received but not yet matched to any invoice, these are identified by their open suspense
  account entry and deducted from the partner's outstanding exposure.
- Bank payments already matched to the invoice, the reconciled amount is read from the invoice's receivable
  line and deducted accordingly.

So with this fix, after a 1,500 bank payment, the system correctly recognises the outstanding amount as 500
and does not show a warning since it is within the 1,000 credit limit.

task-5427613

Forward-Port-Of: odoo/enterprise#119829
Forward-Port-Of: odoo/enterprise#118957

### Issue before this commit:
When reversing an invoice in a Peruvian company, the "Cancel Reason" entered in the credit note window is not propagated to the Peruvian EDI tab of the resulting Credit Note. Only the Credit Reason is successfully reported.

### Steps to reproduce the issue:
1. Download Accounting and l10n_pe
2. Switch to PE company
3. Create an invoice and confirm it
4. Create a credit note for the invoice with a cancel reason and a credit reason and click the reverse button
5. See that in the Peruvian EDI tab only the Credit Reason is reported but not the Cancel Reason

### Cause of the issue:
In the l10n_pe_edi module, the override of the _prepare_default_reversal method maps the l10n_pe_edi_refund_reason to the new move's values, but completely omits the mapping of the wizard's textual reason field to the l10n_pe_edi_cancel_reason field of the resulting credit note.

### Reason to introduce the fix:
To ensure the generated credit notes contain all required information for the Peruvian EDI (SUNAT). Mapping the cancel reason guarantees that the electronic document accurately reflects both the refund code and the descriptive cancellation text provided by the user.

opw-6238525

Forward-Port-Of: odoo/enterprise#119610
Forward-Port-Of: odoo/enterprise#118479

[FIX] voip: make sure any create/write on voip.call syncs user presence
    
    Commit [1] introduced a "in-call" presence icon. Before this commit,
    code updating call records had to call a specific function if user
    presence potentially had to be changed after the record update. While
    not hacking create/write to do that might be prettier, it is also
    subject to mistakes and one was already made: demo data call record
    creation did not update user presence properly. Commit [2] indeed
    introduced calling/ongoing call demo data and the user presence was not
    correct just after database initialization.
    
    This commit fixes that by now potentially syncing in writes and always
    syncing on create.

[FIX] voip: unstuck user call presence sooner in case of stuck calls
    
    At the moment, the Odoo phone has a freshness system for call records
    that appear still calling/ongoing for a strange duration. Indeed, there
    are still cases where a call ended and we could not detect it. For
    example, the user simply closing the tab where a call is ongoing (we try
    to warn the user before he leaves but if agrees to leave anyway, the
    call is just stopped when connexions are lost but the call record stays
    marked as "ongoing"). In those cases, we have 2 things:
    
    - A once-a-month cron checks all cases that are calling for more than
      5 minutes or ongoing for more than 4 hours. It moves them to "ended
      unexpectedly".
    
    - The displayed status in views, shows "calling" / "ongoing" only if the
      record is not older than 5 min / 4 hours. Otherwise it shows "ended
      unexpectedly" already (as if the cron already did its job), despite
      the record still having the "calling" / "ongoing" status.
    
    It is weird and non-perfect but this allows to not have a "heavy" cron
    job and consistent-enough call records display in views.
    
    Of course, the long-term plan is to have more reliable call records
    status (PBX, ...).
    
    A new problem related to this appeared though. Since [1], a call icon
    is used as the discuss presence icon in case the user is currently on a
    call. The "currently on a call" data being transferred based on a field
    "has_active_call" synchronized on call operations. Problem: in the case
    mentioned above (call stuck in ongoing), that field will stay wrong for
    a full month, showing the user as being on call. This commit makes it
    so the check for active calls now considers fresh-enough calls (just
    like the display in views does). It is only updated on a new call
    operation though, so if an user has a stuck call, he will still be shown
    as being on call until he starts/ends another call (or manually correct
    the stuck call record).
    Again, hopefully, stuck call records will be a thing of the past soon
    enough so that issue will be minimized.
    
    Note: this also uses `effective_start_date` instead of `start_date` to
    check for stale calls, as it handles the potential no start_date while
    ongoing that would stay stuck forever (that should not happen but,
    better safe than sorry).

[FIX] voip: not consider incoming calling calls for presence status
    
    Commit [1] introduced a "in-call" presence status. Commit [2], alongside
    several fixes (e.g. parents of this commit and mentioned commit),
    introduced new call demo data, including calling/ongoing calls:
    - One incoming calling for Mitchell Admin
    - One outgoing calling for Marc Demo
    - One incoming ongoing for Marc Demo
    - One outgoing ongoing for Marc Demo
    
    Consequence: both Mitchell Admin and Marc Demo always have the "in-call"
    presence icon, which might not be the best for demo. Still nice to test
    VoIP but misleading for the rest.
    
    In the end, we can have the best of both worlds: at the moment Mitchell
    Admin only has an incoming calling call... and actually, that kind of
    situation should not lead to being consider as "in-call". Calling
    someone does, but receiving a call that we are potentially ignoring at
    the moment does not.
    This commit makes it so incoming calling calls are not considered for
    presence anymore, at the same time thus making Mitchell Admin presence
    not impacted by default VoIP demo data.
    
[1] - https://github.com/odoo/enterprise/commit/f1e0c41fa7b4f425e45303e5e912bf093c56480a
[2] - https://github.com/odoo/enterprise/commit/f16faa029220ca7152289180c4de78783bab03be
    
task-6239844

Forward-Port-Of: odoo/enterprise#118645

### Issue
Since the introduction of the Tax Returns feature in 18.3, it was no longer possible to generate a Closing Entry as the button has been replaced by Returns
The Return mechanism implies that you need a Return Type in order to make a Closing Entry using the Validate button

Additionally, using the Generic Tax Report by default creates a risk in Multi-VAT environments, as it mixes taxes from all countries instead of isolating Peruvian taxes

### Cause
The new 18.3 accounting workflow requires at least one active Return Type associated with a country-specific report variant to display the Return options and process the closing entry Peru was relying on the Generic Tax Report, without a dedicated report variant
No Return Type was configured, which blocked Odoo's automatic VAT closing workflow and prevented the system from prompting the user to configure the required closing accounts

### Steps to reproduce
- Install `l10n_pe_reports` and `accountant`
- Switch to a PE Company
- Go to the Tax Report

Before the fix, no Returns button is available for any of the existing reports, making it impossible to use Odoo's automatic process to configure the tax accounts and trigger the closing entry

### Notes
This is fixed by creating a dedicated Peruvian tax report variant directly in Enterprise that inherits from the generic tax report A custom handler is added to force the domain filtering on Peruvian taxes only, and a corresponding Return Type is defined to restore the full closing entry process safely

opw-5978673

Forward-Port-Of: odoo/enterprise#117891

### Issue:
In DATEV customer and supplier exports, partners outside the European Union still had the `EU-Land` and `EU-UStID` fields filled

However, these fields must only be used for EU countries

For non-EU countries, the `Land` field should be filled instead, and is required whenever the country is not Germany https://developer.datev.de/en/file-format/details/datev-format/format-description/debitorskreditors

### Cause:
`_l10n_de_datev_get_partner_list` did not distinguish between EU and non-EU countries

As a result, any partner with a VAT number could populate `EU-Land` and `EU-UStID`, even if the country was outside the EU

Greece also requires a special case: its VAT prefix is `EL` so the `EU-Land` too, while the country code used in `Land` must remain `GR`

### Steps to reproduce:
- Install `l10n_de_reports` and switch to the DE company
- Create a customer in Switzerland with a valid VAT number
- Create and confirm an invoice for that customer
- Go to Accounting → Audit Reports → General Ledger
- Select the full year
- From the gear menu, export DATEV DATA (zip)
- Open the `EXTF_customer_accounts` file

### Before the fix:
`EU-Land` and `EU-UStID` are filled for the Swiss customer, while `Land` is empty

### After the fix:
`EU-Land` and `EU-UStID` are empty for non-EU countries such as Switzerland, while `Land` is correctly filled

`Land` is filled using the following priority:
1. Partner country_code
2. Country extracted from the VAT number
3. Empty

opw-5902565

Forward-Port-Of: odoo/enterprise#119780
Forward-Port-Of: odoo/enterprise#113835

Currently, an error occurs when user tries to select a property field in auto field.

Steps to replicate:
- Install `sale_management` and `sign`.
- Open Sales > Products > Products > Open any product.
- From the Gear icon, Click Edit Properties and save the record.
- Enable Debug mode if you are using a version lower than 19.0 .
- Open Sign > Configuration > Field Types.
- Create a new Field > Give a name > Select model as `Product`.
- Select Field as `Property > Property 1` and click save.

Error:
- saas-18.3 and later:
```
  File '/home/odoo/odoo18/enterprise/sign/models/sign_item_type.py', line 57, in _check_auto_field_exists
    auto_field_value = record.mapped(sign_type.auto_field)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File '/home/odoo/odoo18/community/odoo/orm/models.py', line 5472, in mapped
    field = records._fields[field_name]
            ^^^^^^^^^^^^^^^
AttributeError: 'Property' object has no attribute '_fields'. Did you mean: 'field'?
```

- saas-18.2:
```
  File '/home/odoo/odoo18/enterprise/sign/models/sign_item_type.py, line 41, in _check_auto_field_exists
    auto_field_value = record.mapped(sign_type.auto_field)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File '/home/odoo/odoo18/community/odoo/orm/models.py', line 5744, in mapped
    if len(records) > PREFETCH_MAX:
       ^^^^^^^^^^^^
TypeError: object of type 'bool' has no len()
```

Cause:
- As the user gave auto fill field as a Property field the [line] called `mapped()` to access its value, this caused the error to occur.

- This occurs because `mapped()` expects a `recordset` (models.Model), but instead it receives a Property object, which does not have `_fields`.

Solution:
- Using `'allow_properties': 'False'`, the property fields wont appear in the list of field selection.

[line]: https://github.com/odoo/enterprise/blob/cdaeb79e1f623831fffa553dbb658698367c7e19/sign/models/sign_item_type.py#L41

sentry-7378769090

Forward-Port-Of: odoo/enterprise#119641
Forward-Port-Of: odoo/enterprise#113091

### **Steps to reproduce:**
- Install Payroll and Attendance apps.
- Create an employee with a flexible working schedule and work entry source as attendance.
- Create an attendance record for this employee.
- Create and compute a payslip for this employee.

### **Observed Behavior:**
 Worked Day lines are not generated, and Basic Wage is calculated as 0.

### **Expected Behavior:**
Worked Day lines should be populated based on attendance records.

### **Root Cause:**
During payslip computation, [_compute_worked_days_line_ids](https://github.com/odoo/enterprise/blob/4339010eb1e1633a67573d08e032f3922b0bec49/hr_payroll/models/hr_payslip.py#L1846) only generated work entries for versions having a `resource_calendar_id`  at [1]. As a result, fully flexible employees without a working schedule were excluded from work entry generation, preventing worked day lines from being computed.

[1]- https://github.com/odoo/enterprise/blob/4339010eb1e1633a67573d08e032f3922b0bec49/hr_payroll/models/hr_payslip.py#L1890-L1898

### **Fix:**
Remove the `resource_calendar_id` filter when calling `generate_work_entries` in `_compute_worked_days_line_ids` so work entries are also generated for fully flexible employees using attendance-based work entries.

**opw-6146452**

Forward-Port-Of: odoo/enterprise#119673
Forward-Port-Of: odoo/enterprise#117409

## Issue
When scheduling a sign request, the request appears immediately on the portal for the requested signer.

## Steps to reproduce
1. Install *Sign* (`sign`)
2. Create and send a sign request
    - Signer 1: Any portal user (e.g., Joel Willis)
    - Use the clock icon to schedule the signature request to a future date
3. Log in as the portal user used in step 2
4. Navigate to Signature Requests
5. **The signature request already appears in the list, even though it was scheduled for a future date.**

## Cause
The portal filters the sign requests shown based on the `is_mail_sent` field, which does not properly reflect when the signature request is shared to the user.

https://github.com/odoo/enterprise/blob/9898272f17809decf6c5bb4600aca46f9ffae6f0/sign/controllers/portal.py#L40

In fact, when scheduling a signature request, the `is_mail_sent` field is unconditionally set to `True`, even if the signature request will only be sent later.

https://github.com/odoo/enterprise/blob/9898272f17809decf6c5bb4600aca46f9ffae6f0/sign/models/sign_request_item.py#L289

## Fix
Since the `"scheduled"` `sign_request_item.state` option introduced by https://github.com/odoo/enterprise/commit/ed8d5a653e01b1378f0020e2f7a7c2d39fadf3e9 in 19.1, we can easily filter out the sign request items that are scheduled. That state is automatically updated by the `_cron_update_state`, introduced by the same commit as the `"scheduled"` option.

https://github.com/odoo/enterprise/blob/9898272f17809decf6c5bb4600aca46f9ffae6f0/sign/models/sign_request.py#L493-L503

opw-6227472

Forward-Port-Of: odoo/enterprise#119007
Forward-Port-Of: odoo/enterprise#118491

Currently, when a user deletes a line and validates internal movement in the barcode system, the system allows validation even though specifying the destination location after each scan is required.

## Steps to produce:
- Install the Inventory module
- Go to Settings and enable Storage Locations.
- Inventory > Configuration > Operation Types > Internal Transfers > Barcode App
- Configure the Destination Location to require scanning after each product.
- Create an Internal Transfer for Pedal Bin, demand 1.
- Mark the transfer as To Do and open it in the Barcode app.
- Add quantity using +1, then scan the barcode for the Pedal Bin(Barcode: 6016478556493).
- Delete the newly added line and attempt to Validate.

## Observed Behavior:
The system should prevent transfer validation when the destination location has not been scanned and display a notification to the user, similar to the behavior before user deleted the newly added line.

## Root cause:
This issue occurs because when the delete button is pressed, the deleteLine function [1] removes the line, but the deleted line becomes the selected line due to [2] being triggered before the UI updates.

As a result, the selected line is now undefined. Since the selected line is undefined, it fails to meet the condition at [3] during validation. This prevents notifications from being triggered and allows the transfer to be validated before the destination location has been scanned.
[1]:
https://github.com/odoo/enterprise/blob/3476d15bf8e75eb6530658dd623861b60963ab40/stock_barcode/static/src/models/barcode_model.js#L826-L836 [2] :
https://github.com/odoo/enterprise/blob/327d4478128f33fb2e0c477533bd4983178abf17/stock_barcode/static/src/components/line.js#L129-L133 [3]:
https://github.com/odoo/enterprise/blob/6ff158ca3a6d2d2b3d285a7f8317622844811688/stock_barcode/static/src/models/barcode_picking_model.js#L945-L948

## Solution:
We can prevent users from validating if any line has an unscanned destination location when destination-location scanning is mandatory after scanning each product.

To enforce this behavior, we can track whether a line has been modified and whether a destination location has been scanned and applied to that line. This allows us to identify which lines still require destination location scanning before validation can proceed.

However, line state information is currently discarded and recreated on every save. As a result, information about lines that were updated and already had their destination location scanned is lost. This may incorrectly require users to rescan the destination location, even though it was previously scanned.

To address this, we preserve the destination-scanned and modified state by carrying it forward from existing lines to their corresponding newly created versions using a loop. This ensures that destination location scan status is retained and users are not asked to rescan unnecessarily.

opw-6069614

Forward-Port-Of: odoo/enterprise#119870
Forward-Port-Of: odoo/enterprise#113618

### Issue before this commit:
In accrual reports (e.g., "Bill To Receive", "Billed Not Received", "Invoices To Be Issued", and "Invoices Not Delivered"), when grouping the list view by fields such as Vendor, the group header totals for the "Received", "Billed", and "Amount" columns display 0.00 even if the interanl lines of the group are not 0.00.

### Steps to reproduce the issue:
1. Download Purchase Accounting and Sale Accounting
2. Go to one of this pages: Billed Not Received, Bill To Receive, Invoices To Be Issued, and Invoices Not Delivered
3. Ensure the view is in its default grouping (grouped by Vendor or Customer)
4. Observe the group header rows for the Received (or Delivered), Billed (or Invoiced), and Amount columns. They all display 0.00
5. Expand a group that contains records with values greater than zero
6. Observe that the individual records populate correctly, but the aggregated group header row continues to display 0.00.

### Cause of the issue:
The commit ddc1b681656ea8c70f3231cda20b5a58b9ff7dd6 adapted the code to retrieve the new accrual reports but attempted to fetch grouped records using group[0].id as the dictionary key, while the grouped() method actually used the recordset object as the key. This mismatch caused the dictionary lookup to fail, resulting in 0.00 sums. https://github.com/odoo/enterprise/blob/c8535a7a0e2eae811048a34a0bae187a1fa45311/account_accountant/models/analytic_mixin.py#L40-L48

### Reason to introduce the fix:
This fix restores the core analytical utility of the accrual reports, which are crucial for accountants during period-end closings to evaluate totals at a glance.

opw-6232273

Forward-Port-Of: odoo/enterprise#118399

Steps to Reproduce
---
1. Open the Planning module
2. Create an open shift with allocated_hours
3. Click Auto Plan
4. Click Undo on the auto-plan notification

Issue
---
Undoing an auto-plan operation triggers recomputation of allocated_hours,
causing the shift to lose its original workload value.

Current Behaviour
---
When resource_id is set to False during undo:
- _compute_allocated_hours is triggered (depends on resource_id)
- _compute_allocated_percentage is triggered (depends on allocated_hours)
- Both fields are recalculated, potentially changing allocated_hours
  from its pre-assignment value

Expected Behaviour
---
Undoing auto-plan should preserve allocated_hours at its pre-assignment
value while allowing allocated_percentage to adapt to the new context
(open slot vs assigned resource).

Fix
---
Use protecting context manager in action_rollback_auto_plan_ids
to prevent allocated_hours from being recomputed when resource_id is
removed. This allows allocated_percentage to recalculate naturally based
on slot duration while keeping allocated_hours stable.

task - 4952149

Forward-Port-Of: odoo/enterprise#119941
Forward-Port-Of: odoo/enterprise#102864