fix(acf): a flexible content row is read with its own layout; a key and name for one sub-field is reported

[ABB65]

Closes #45.

Label: GENERIC. This is ACF/SCF flexible content and keyed-row handling in Source::acf_value, used by every site whose flexible layouts reuse a sub-field name. It is not tied to one builder or theme.

What changes (includes/class-contentrain-bridge-source.php)

1. Layout from acf_fc_layout. Before, the sub-fields of every layout were merged into one lookup by name, so a name that two layouts share resolved to the last layout's definition for every row. That could mean the wrong type, or a field judged secret, or one that wasn't. Now a row whose acf_fc_layout names a known layout looks its cells up in that layout first. Any cell that layout does not define (for example another layout's field key), and every cell of a row with no or an unknown layout, falls back to the all-layouts lookup main uses, so it is still typed and a secret type is still removed (fail-closed; 5515861, review follow-up).
2. Key + name collision. A row holding one sub-field under both its key (field_…) and its name kept the later cell silently. It still keeps the later cell, so output is unchanged, and it now reports {source: <path>/<name>, reason: 'acf-key-name-collision: kept <cell>, dropped <cell>'} in the excluded list, next to acf-key-unmapped, using the existing reason: detail convention.

Tests (tests/integration.php, 2 new checks)

• Two layouts both define media, one as image and one as password. The hero row keeps media: 42 and the note row's media is removed as sensitive. With the old union lookup the hero value was dropped too, so this check fails on main.
• A repeater row with field_soc_label and label keeps the later cell and reports exactly one acf-key-name-collision: kept label, dropped field_soc_label.
• A hero row carrying field_note_media (the note layout's password key) with the value hunter2: the cell is removed as sensitive-field, the value appears nowhere in the output, and no acf-key-unmapped is reported. This check fails against the layout-only lookup of the first commit (verified by running the suite on that code).

Checks (local, clean stack: down -v, then KEEP=1)

• npm run test:wordpress: 397 checks passed (main has 394, plus these 3)
• npm test: 13/13
• npm run test:reader: pass
• npm run test:plugin: Plugin Check 0 errors, 4 warnings (the four known WPML hook names)
• php -l on all 59 PHP files: no errors

No version bump or changelog entry here. Add them at the next release.

[ABB65]

ONAY (t8) @ 5515861 — lean review, per the founder's "product first" order.

• Code read: Source::acf_value looks up the row's own layout (acf_fc_layout) first, then falls back to the lookup across every layout, so a cell from another layout is still typed and a secret type is still removed (fail-closed). The key+name collision is reported with the existing reason: detail convention, and output is unchanged.
• The 3 new tests cover the shared name, the collision and the planted key from another layout.
• CI green at this head: acceptance (test:wordpress), small, small (10), plugin metadata. merge-tree vs origin/main: clean. php is not installed here, so CI is the lint/test evidence.
• Non-blocking note: when a cell from another layout resolves to the same name, as in the planted test (media + field_note_media in a hero row), the report says kept field_note_media, dropped media. In reality media: 42 is kept and field_note_media is removed as sensitive. The "later cell wins" wording is also wrong when the later cell is cleaned to null. A message that names the cell actually kept could come in a later PR.