### Problem

When an AI agent includes a document source whose underlying `ir.attachment`
has `res_field` set (e.g., an invoice PDF generated from a Sale Order),
triggering **Update With AI** from **Website → Site → Optimize SEO**
raises a `KeyError` in `_build_rag_context`.

### Steps to Reproduce

1. Open the **AI** app.
2. Configure the **Odoo Agent**.
3. Add a source → **Add From Documents**.
4. Select a document whose underlying `ir.attachment` has `res_field` set
   (e.g., an invoice PDF generated from a Sale Order).
5. Go to **Website → Site → Optimize SEO**.
6. Click **Update With AI**.
7. Observe the following error:

```
KeyError: 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'
```

[Video](https://drive.google.com/file/d/1AAg0AvC3TRQzMLhI9hWFuSarOTHzQE-b/view?usp=sharing)

### Root Cause

In `ai/models/ai_agent.py`, `_build_rag_context()` retrieves the
`ai.agent.source` records corresponding to the embeddings by searching on the
attachment checksum:

```python
agent_sources = self.env["ai.agent.source"].search([
    ("attachment_id.checksum", "in", embeddings_attachment_checksums),
    ("agent_id", "=", self.id),
])
```

This domain traverses `attachment_id.checksum`, which internally calls
`ir.attachment._search()`.

As part of the standard attachment search behavior,
`ir.attachment._search()` automatically injects a
`('res_field', '=', False)` filter unless:

- `skip_res_field_check=True` is set in the context,
- the domain explicitly references `id` or `res_field`, or
- `bypass_access` is enabled.

```python
domain = Domain(domain)
if (
    not self.env.context.get("skip_res_field_check")
    and not any(d.field_expr in ("id", "res_field") for d in domain.iter_conditions())
    and not bypass_access
):
    disable_binary_fields_attachments = True
    domain &= Domain("res_field", "=", False)
```

[Reference](https://github.com/odoo/odoo/blob/19.0/odoo/addons/base/models/ir_attachment.py#L630)

Because of this implicit filter, attachments with `res_field` set are excluded
from the search. Consequently, `agent_sources` does not contain all the sources
corresponding to the retrieved embeddings.

Later, `_build_rag_context()` builds a checksum-to-source mapping:

```python
source_map = {
    source.attachment_id.checksum: source
    for source in agent_sources
}

for embedding in similar_embeddings:
    checksum = embedding.attachment_id.checksum
    agent_source = source_map[checksum]
```

Since `source_map` is built from the incomplete `agent_sources` recordset, it is
missing entries for attachments filtered by `ir.attachment._search()`.
However, `similar_embeddings` still contains embeddings for those attachments.

As a result, the lookup:

```python
agent_source = source_map[checksum]
```

raises a `KeyError`.

### Solution

Bypass the implicit `res_field` filter when searching `ai.agent.source`:

```python
agent_sources = (
    self.env["ai.agent.source"]
    .with_context(skip_res_field_check=True)
    .search([
        ("attachment_id.checksum", "in", embeddings_attachment_checksums),
        ("agent_id", "=", self.id),
    ])
)
```

This ensures that all `ai.agent.source` records matching the requested
attachment checksums are returned, including those referencing attachments with
`res_field` set. As a result, `source_map` contains all expected entries and
`_build_rag_context()` no longer raises a `KeyError`.

opw-6379816

Forward-Port-Of: odoo/enterprise#125780

Problem:
On a flexible working schedule, a time off of a few hours (neither a full nor a half day) was treated as a full day off. The Attendance list then reported the whole day's attendance as overtime, and the Time Off gantt grayed out the entire day instead of only the leave's hours.

Steps to reproduce:
1. Give an employee a flexible working schedule (e.g. 8h/day) with an overtime ruleset based on the contract's expected hours.
2. Record a 2-hour time off, then an 8-hour attendance on the same day.
3. Observe the attendance reports 8 hours of overtime instead of 2.

Current behavior:
A partial time off makes the whole day count as extra hours.

Expected behavior:
Only the hours actually taken off reduce the day's expected hours.

Cause:
For a flexible schedule, _handle_flexible_leave_interval expands a leave to the whole day. The override already narrows full-day and half-day leaves, but any other number of hours fell through to that full-day expansion. The expanded interval is subtracted from the day's expected hours in _work_intervals_batch, so they drop to zero and every worked hour becomes overtime.

Fix:
A leave of an arbitrary number of hours should only remove the hours it actually covers, so it keeps its requested interval instead of being stretched to the whole day.

opw-6291536

Forward-Port-Of: odoo/enterprise#127257
Forward-Port-Of: odoo/enterprise#123778

Steps to reproduce:
-------------------
1. Install `l10n_mx_edi_pos` and set the company to Mexico.
2. In PoS, make an order with one product and pay it.
3. Open the order in the backend, click "Return Products" to make a refund, but don't pay it, cancel it instead.
4. From the same order, click "Return Products" again to make a second refund, and pay it normally.
5. Go to the orders list, select the main order and the paid refund (not the cancelled one), then Actions > Create Global Invoice.

-> Observation: error in the global invoice.

In the CFDI tab of the main order the line is "Send Global In Error", and hovering on it the detail says "Failed to distribute some negative lines".

Why:
----
When we make the global invoice, we remove the refunds from the order. A cancelled refund was never paid, so we should not count it. But we were counting it too. So we removed the refund amount twice in our case, one for the paid refund, and one for the cancelled one, and we end up with an order with negative amount that cannot be distributed.

The fix:
--------
We now skip the cancelled orders when we search the refunds, the same way it is done above when we collect the refunded orders.

opw-6261404

Forward-Port-Of: odoo/enterprise#127724
Forward-Port-Of: odoo/enterprise#120996

**Steps to reproduce:**
- Go to Documents app in mobile
- Go to the kanban view
- Add some files and select one
- Click on `Info & Tags` button in the control panel
- Reload the page
- Chatter is not displayed but the button is still enabled
- Switching to the list view properly shows it

**Issue:**
Original fix (see [1]) was not enough for every case. Additional issues:
- Chatter hidden on init even when its panel has `visible = true`
- State desynchronized with the view when switching menu type (kanban/list) or by previewing a document and coming back
- When using the button with an open preview, chatter shows up in the background but is not accessible (and going back discards it)
- Removing selection with an open chatter disable the related action

**Fix:**
- Disable the chatter on mobile init by default to avoid having to manually move it back
- Reset chatter on selection removal to avoid getting stuck in the menu
- Reset chatter on view switch to avoid being in the wrong state afterwards
(and revert the previous css changes)

Not a great fix (quite mobile-specific) and there might still be some edge cases.

[1] original fix: https://github.com/odoo/enterprise/commit/cf3c2fce8a6b7b2d7547d44a0e4423f887986d52

opw-6061993

Forward-Port-Of: odoo/enterprise#121521

…ern pypdf backend

`export_article_to_pdf` calls `writer.setPageMode("/UseOutlines")` on the `PdfFileWriter` instance to make the outline/bookmarks panel visible by default when the generated PDF is opened. the old PyPDF2

`odoo.tools.pdf` picks its backend dynamically it first tries to import PyPDF2==2.12.1  and only falls back to the modern `pypdf` library if PyPDF2 is not importable, which is the case on Python 3.13 per requirements.txt, or on any worker where PyPDF2 failed to install.

The PyPDF2-based writer still exposes `setPageMode`, so the bug was never seen on backends using it. The modern pypdf-based writer replaced that method with a `page_mode` property (getter/setter) and never kept a camelCase alias for it

As a result, any request hitting this code got a beautiful HTTP 500
    `AttributeError: 'BrandedFileWriter' object has no attribute 'setPageMode'`

`setPageMode` is not called anywhere else so
this fixes the call site directly instead of adding a new alias to the shared _pypdf.py

opw-6382905

Forward-Port-Of: odoo/enterprise#124480

Steps to reproduce:
-------------------------------------------------
1. Install the `sale_project` module
2. Create a test user with Project User rights
3. Create a Project Role with the Created User as a Team Member
4. Create a Template Project as follows:
    * Add one task to the template project
    * Add the created Project role to the Task
5. Create a Service Type Product with:
    * Create on order: Project
    * Project Template: Created Template
6. Archive the Created User
7. Create and Confirm the Sale Order with the Created Product

Observation:
-------------------------------------------------
The generated task is assigned to the archived user, although the archived user is no longer part of the Project Role.

Issue:
-------------------------------------------------
While creating Project and Tasks from template, the context disable active record filtering (e.g., `active_test=False`), causing the assignment logic to fetch both active and inactive/archived users linked to the role. https://github.com/odoo/odoo/blob/8ec646e51497b38d34ea59296e0fc8644a50ee3a/odoo/orm/models.py#L4868

After that, during the `copy_data` method, It takes all the users from the roles without checking weather user is active or not
https://github.com/odoo/odoo/blob/8ec646e51497b38d34ea59296e0fc8644a50ee3a/addons/project/models/project_task.py#L890-L904

And even if we pass only Active users from this method, on moving further, it reassigns the users from roles without checking the Active field of the user

https://github.com/odoo/enterprise/blob/5abb147f9bf725daafc202d8259a5bb8a9b78d94/project_enterprise/models/project_task.py#L501-L503

https://github.com/odoo/enterprise/blob/5abb147f9bf725daafc202d8259a5bb8a9b78d94/project_enterprise/models/project_task.py#L544-L553

Due to this, the Archived User is also assigned to the tasks from the project roles

Solution:
-------------------------------------------------
Apply a `filtered('active')` check directly on the project role's users `(role.user_ids)` within the core task-copying logic in both `project` and `project_enterprise` modules. This ensures archived users are universally excluded from task assignments during template copying, regardless of what triggers the template instantiation.

Related Community PR: https://github.com/odoo/odoo/pull/274426

opw-6350841

Forward-Port-Of: odoo/enterprise#127529
Forward-Port-Of: odoo/enterprise#125637