[20.0][MIG] document_page: Migration to 20.0

[dnplkndll]

Port of document_page from 19.0 to 20.0. Follows the OCA migration guide.

Depends on #646, so Detect unreleased dependencies is red until that lands.

ir.model.access and ir.rule are both removed in 20.0 — the security file is converted to ir.access by upgrade_code --script 19.4-00-ir-access, with the two global rules becoming rows with an empty group_id.

AI-assisted (Claude Code); every change reviewed, tested, and owned by the author.

[XtremXpert]

Thanks for the migration! While running document_page on 20.0 we hit a bug that this branch still carries (the same one exists on 19.0, see #649):

After editing a page's content, history_head keeps pointing to the previous revision, so the page shows the old content on reload and the edit looks lost. _compute_history_head takes history_ids[0], relying on _order = "id DESC", but in the cache the revision created by _inverse_content() during write() is appended at the end of history_ids.

Fix (same as #649):

    def _compute_history_head(self):
        # history_ids is not sorted by _order in cache: a revision created
        # during write() comes last, so pick the newest one explicitly.
        for rec in self:
            rec.history_head = rec.history_ids.sorted("id", reverse=True)[:1]

with a regression test, test_page_edit_keeps_new_content (create a page, write new content, flush and invalidate, then check the head and the content). It's in our 20.0 fork: XtremXpert@118e02e. Feel free to cherry-pick it, or I can open a PR against your branch if you prefer.

[XtremXpert]

Thanks for the migration! One issue we hit running document_page on 20.0: after editing a page, history_head points to the previous revision, so the saved content looks lost. Two suggestions below (fix + regression test); we run them in production on 20.0.

[dnplkndll]

**XtremXpert **

@XtremXpert looks good thanks