…heet history

Some spreadsheet can end up in a corrupted history state

Original Data ⮕ no revision (XX) ⮕ Snapshot ⮕ Active Revisions and where the original data differ from the snapshot.

XX: can occur after the script added https://github.com/odoo/upgrade/issues/6340. The migration script is supposed to rewrite the history of the spreadsheet when there are holes in the archived revisions continuity (eg. the history was original Data ⮕ rev **A** ⮕ rev **B** ⮕ rev **C** ⮕ snapshot ⮕ rev **D** and user deleted rev **B** for instance, the script deletes **A** and **C** and marks **D** as the very first revision).

Version history was designed with the idea to replay every single revision that existed since the creation of the spreadsheet and apply it to the original data, and in case of missing revisions, in our example, B is missing, we detect the lack of continuity, we start from the snapshot, and replay every single revision since that snapshot.

When the migration fixes the continuity, by deleting all the old revisions, and changing their order, we can no longer detect the lack of continuity and end up replay every available revision to the original data ,even though they are based on the snapshot.

In that scenario, since the revisions will be applied on the original data but since they are based on the state of the snapshot only, the final state of the spreadsheet will be corrupted since a part of the . If a user then decides to restore the spreadsheet to the version they see in the version history (which is corrupted as mentioned) and the last snapshot is overwritten with the corrupted state, effectively breaking the spreadsheet.

With this revision, we add a detection of this corrupted history state, in which case we enforce the history to be replayed from the snapshot and not the original data.

Task-6533708

Forward-Port-Of: odoo/enterprise#131803
Forward-Port-Of: odoo/enterprise#130649

### Problem

`l10n_ar_edi` reports WSFEX / WSBFE line quantities (`Pro_qty`) truncated to 2 decimals in 19.0, no matter what precision the database has configured.

`_get_line_details()` reads the precision like this:

```python
uom_precision_digits = min(self.env['decimal.precision'].precision_get('Product Unit of Measure'), 2)
```

In 19.0 that `decimal.precision` record was renamed to **`Product Unit`** (`uom/data/uom_data.xml`, `decimal_product_uom`), so the lookup matches no record. `precision_get` does not raise in that case, it falls back to a default of 2 (`base/models/decimal_precision.py`: `return res[0] if res else 2`), so `min(2, 2) = 2`.

The rename was applied everywhere else — `account.move.line.quantity` already declares `digits='Product Unit'`. Only this lookup kept the old name.

### Impact

A quantity of `0.042` is sent as `0.04`. ARCA rejects the invoice with **error 1815** (item math inconsistency: `unit price x quantity - discount` no longer matches the item total), which blocks export invoicing completely.

It only shows up after upgrading to 19.0: on 18.0 the record still had the old name, so the lookup worked and the configured precision was used.

On our hosted fleet we identified **67 Argentinian databases** that use WSFEX or WSBFE and have a unit precision above 2. Eight of them already run 19.0 and are affected today; the rest will hit the same rejection as they upgrade.

### Fix

Use the current record name, and raise the cap to **6**, the maximum `Pro_qty` accepts according to the WSFEX developer manual ([V3.1.1](https://www.afip.gob.ar/ws/documentacion/manuales/WSFEX-Manualparaeldesarrollador_V3.1.1_ARCA.pdf), p. 15).

### Test plan

`l10n_ar_edi/tests/test_fex.py` adds `test_ar_edi_wsfex_pro_qty_decimal_precision`, covering three cases:

| `Product Unit` | quantity | expected `Pro_qty` |
|---|---|---|
| 3 | 0.042 | `0.042` |
| 2 | 0.042 | `0.04` |
| 8 | 0.1234567 | `0.123457` (capped at 6) |

It does not go through the ARCA mock: it calls `_get_rounded_base_and_tax_lines()` and `_get_line_details()` directly.

### Note

Supersedes #130582 by the same author, which carried the same one-line fix without a test. Please review this one instead.

### Left out on purpose

`price_precision_digits` in the same method caps the unit price at 3 digits, which does not come from the specification either. That lookup still uses a record name that exists (`Product Price`), so there is no regression, and we have no rejection reported because of it. Changing it would widen the scope of a fix that has a concrete incident behind it, so it is left untouched here.

Forward-Port-Of: odoo/enterprise#131457