Audit the other #340 catalogues for the formal/informal register fault found in Hungarian

[dannymcc]

Issue #350 found that the bulk translation fill in #340 introduced formal-register strings into the Hungarian catalogue, which had been consistently informal since well before #340. Twenty-eight msgstrs had to be rewritten in 4c14615 to put it back.

The same fill produced the other eighteen catalogues, and several of those languages draw the same T–V distinction: German (Sie/du), French (vous/tu), Spanish (usted/tú), Italian, Portuguese, Dutch, Polish, Czech, Russian among them. There is no reason to think Hungarian was uniquely affected, and we only caught it because a native speaker raised it.

Scope: an audit, not a blanket rewrite. For each affected catalogue under app/translations/*/LC_MESSAGES/messages.po:

1. Establish the settled register from the catalogue's own history — the strings that predate the Still hardcoded or missing keys for translation #340 fill (git log around commit 2132f12) are the evidence, exactly as they were for Hungarian. Do not impose a house style where the catalogue has not already chosen one.
2. Compare the strings Still hardcoded or missing keys for translation #340 added against that baseline and list the divergences.
3. Where a catalogue has a clear pre-Still hardcoded or missing keys for translation #340 register and Still hardcoded or missing keys for translation #340 broke it, correct the offending msgstrs — register only, no other wording changes.
4. Where the pre-Still hardcoded or missing keys for translation #340 content is too thin or genuinely mixed to establish a baseline, record that and leave it alone. That is a question for a native speaker, not for us.

Start with de, fr and es: they are the largest catalogues and the most likely to be read.

Acceptance criteria:

• A written finding per catalogue: settled register, or "no baseline, left alone" with reasoning.
• Corrections landed only for catalogues with an established baseline, register-only, msgid set still matching messages.pot exactly, nothing left blank or fuzzy.
• A guard test in the shape of tests/test_i18n_issue_350.py for each catalogue that gets corrected, so the register cannot silently regress on the next bulk fill.
• The full suite passes.

Hungarian is already done (#350) and is out of scope here.

[dannymcc]

Thanks for flagging this — worth checking properly given how #350 came about.

I've audited all nine catalogues you named (de, fr, es, it, pt, nl, pl, cs, ru) against their own history: pulled each one as it stood immediately before the #340 fill (git show 2132f12^:app/translations/<code>/LC_MESSAGES/messages.po) as the baseline, then checked what #340 actually added (git diff 2132f12^ 2132f12 -- app/translations/<code>/...) against it. For the pro-drop languages (Czech, Russian, and Polish, which — like Hungarian — mark register mainly through verb conjugation rather than a standing pronoun) I checked the imperative verb forms directly, not just possessive pronouns, since a pronoun-only check would have missed exactly the kind of thing that went wrong in Hungarian.

The result is reassuring: I didn't find the Hungarian-style break in any of the nine.

• de: formal (Sie/Ihr) throughout, unchanged by Still hardcoded or missing keys for translation #340.
• fr: formal (vous/votre) throughout.
• es: formal (usted/su — "Compruebe", "Ponga") throughout.
• nl: formal (u/uw) throughout, no "je/jouw" anywhere.
• cs: formal, via the -ejte imperative ("Zadejte", "Přidejte") and váš/vaše — consistent before and after.
• ru: formal, via the -ите imperative ("Укажите", "Добавьте") and ваш — consistent before and after.
• it: informal (tu — "Assegna", "Aggiungi", "Scegli"), no formal Lei-forms anywhere.
• pl: informal (ty/Ci), and it already has its own convention of capitalising the possessive ("Twój/Twoje/Ci") as a politeness marker — Still hardcoded or missing keys for translation #340's additions ("Twoje konto...") follow that convention rather than breaking it. I flagged this to myself as a possible drift at first and it turned out to be a false alarm once I checked the pre-Still hardcoded or missing keys for translation #340 baseline properly.
• pt: this catalogue is Brazilian Portuguese ("usuário", "Registro"), and consistently uses the "você" register throughout (3rd-person subjunctive imperatives like "Adicione", "Dê um nome", plus seu/sua) — no "tu"-conjugated forms anywhere, before or after Still hardcoded or missing keys for translation #340.

So: nothing to correct here, and no guard tests are needed under the acceptance criteria as written, since that's tied to catalogues that actually get corrected. Hungarian looks like an isolated slip in that one batch rather than a systemic problem with the fill.

If it's useful anyway, I can add a small pinning test (tests/test_i18n_issue_351.py) recording the register markers found above for each catalogue, purely so a future bulk fill that reintroduces drift gets caught by CI rather than by a native speaker after the fact — let me know if you'd like that added, otherwise I'd suggest closing this one out. I'd also want a full manual read of the it and pl diffs rather than my sampling before calling it fully done, since those two got spot-checked rather than grepped exhaustively the way de/fr/es/nl/cs/ru were.

[dannymcc]

A fix for this has been pushed to dev and will be included in the next release.

Audited the nine T-V catalogues from #340 (de, fr, es, it, pt, nl, pl, cs, ru) against their pre-#340 baseline and found no register break in any of them, confirming the triage plan. Method: loaded each catalogue at 2132f12^ and now with babel, split msgids into baseline versus #340-added, then checked address pronouns and — for the pro-drop languages, where the verb carries the register — imperative conjugation on every added string of four words or more. de/fr/es/nl/cs/ru are consistently formal, it/pl consistently informal, pt consistently você-register; all nine match messages.pot exactly with nothing blank or fuzzy. No .po file was touched. Added tests/test_i18n_issue_351.py (98 tests) pinning each catalogue's settled register: no wrong-register marker anywhere, a floor on how many strings still carry the right one, and the pot/blank/fuzzy checks that previously covered only Hungarian. Each guard was verified to fire by probing it with the #350 fault transposed into that language. Full suite passes: 1375 tests.