Description of the issue/feature this PR addresses:
When grouping datetime fields by hour, `read_group` formats the group display label using `hh:00 dd MMM`.

In Babel/LDML formatting, `hh` represents a 12-hour clock. Since the format does not include an AM/PM marker, afternoon/evening hours are displayed ambiguously in grouped views.

Current behavior before PR:
A datetime value in the afternoon is grouped under a 12-hour label without AM/PM.

For example, records around `13:50` are displayed under:

    01:00 20 Mar

Similarly, a datetime value around `16:20` may be grouped under:

    04:00 26 Mar

This is ambiguous because the group header does not indicate whether the hour is AM or PM.

Example screenshot showing records around 13:xx grouped under `01:00`:
<img width="251" height="195" alt="image" src="https://github.com/user-attachments/assets/9c394063-f679-4faa-a045-2456cc452fb9" />


Desired behavior after PR is merged:
Hour-based datetime group labels should be unambiguous.

The hour grouping format now uses `HH:00 dd MMM`, so grouped datetime labels render using a 24-hour clock.

For example:

    13:00 20 Mar
    16:00 26 Mar

This fixes the datetime hour grouping label shown in grouped list views and other `read_group` consumers.

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

The following use case has been observed:

0. Customer start a payment from /shop/payment.
1. We received the webhook that notifies that the payment succeeded.
2. The payment post-processing cron start (it gather all the transactions that need to be processed, including the customer new transaction)
3. Meanwhile, the customer is redirected back by the payment provider to Odoo, which then redirect to /payment/status and start the payment post-processing for that specific transaction
4. The customer initiated payment processing finishes, he is redirected back to /my/orders/... page.
5. The payment post-processing cron finally start processing the same customer transaction and process it (a second time).

In that case, as the transactions to be post-processed backlog was quite high, there is consequent time between the time we gather all the TXs to post-process and actually process the customer transaction.

Also we don't end up with a `SerializationError` as the cron do commit after each transaction post-processing.

This commit force invalidate individual transaction cache values and recheck if it effectively still need to be post-processed before doing it.

opw-6332192

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

### Steps to Reproduce
1. Create SO with one line. Unit price $100 + 15% tax
2. Create a down payment for $100 and confirm
3. Create a down payment for the remaining $15 and confirm
4. Create an invoice for the original SO, the total is 0.00
5. Download the PDF and notice that the total is -0.00

### Description of the issue/feature this PR addresses: 
**Issue:** There should not be a negative sign on the invoice PDF. Because of rounding to accommodate the limitations of binary memory, there is a negligible negative remainder sometimes, which gets shown as -0.00.

**Solution:** Update the _compute_tax_totals method to check if the calculated totals and subtotals evaluate to zero using the currency's precision (is_zero()). If a value evaluates to zero, we explicitly force the amount to 0.0 and reformat it using formatLang, ensuring the PDF displays a clean 0.00

### Current behavior before PR:
Some cases cause there to be a negative zero on the invoice PDF.

### Desired behavior after PR:
No negative zeroes on the invoice PDF.

opw-6298712

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

The Thread component mirrors `thread.isLoaded` into the `state.mountedAndLoaded` flag that gates, in the template, whether the real messages are rendered. That mirroring effect both read `mountedAndLoaded` as one of its dependencies and wrote it.

`useEffect` records its dependency array before running the body, so right after the effect sets `mountedAndLoaded` to true the recorded dependencies still hold the pre-write `[isLoaded=true, mountedAndLoaded=false]` pair; that update only settles on a later, microtask-deferred patch. When a second reload runs `reset()` in that window it drives `mountedAndLoaded` back to false while `isLoaded` stays true, and the settling patch then computes the very `[true, false]` pair already recorded. The effect never re-runs, so `mountedAndLoaded` is stranded at false and no message is rendered.

Depend on a monotonic `resetCount` instead. Reading it in the effect dependency array subscribes the render to it (OWL subscribes a `useState` proxy's render callback on every read, wherever it happens), so a `reset()` bump re-renders and re-runs the mirror to re-sync `mountedAndLoaded` with `isLoaded`. Bump it only when `isLoaded`: while loading, `applyScroll` resets on every patch, so an unconditional bump would spin the render loop; the guard re-arms only in the case that heals. `reset()` still clears `mountedAndLoaded` (the false dip is needed for the reload scroll handshake), so behaviour is otherwise unchanged.

The race is not deterministically reproducible with this version's test tooling, which cannot advance the render loop a single frame, so no test is added here; the fix is covered by tests on later versions.

https://runbot.odoo.com/odoo/error/940032