Add Finnish (fi) translations - #142

Thanks — pushed 645bc9e. All nine inline threads are answered and resolved; eight taken as suggested, carboy/demijohn half taken (details on that thread).

Partitive singular — you're right, and the label was wrong in two places. "20 tuntia" is singular; "tunteja" would be the plural. Every form in the file was already the right one, but I'd described the rule wrong in both the PR body and the first commit message. PR body is corrected; the commit message I've left alone rather than rewriting pushed history.

Key parity — added a guard. TestLocaleKeyCoverage in locales/locales_test.go checks every locale against the English key set plus the union of currency codes across locales (since en has no currencies.yml, as you noted). Two exemption mechanisms:

• exemptKey — regimes.* and proper nouns like Codice fiscale, which are English-by-design.
• knownGaps — a per-locale list of keys that fall back to English today, all pre-dating the test. A new locale must ship with no entry there, and the test also fails if an entry becomes stale, so the list can only shrink.

I verified it has teeth rather than trusting it: deleting a key from fi/app.yml fails it, and so does a stale knownGaps entry. There's also TestLocaleFilesUseOwnCode, which catches the classic copy-a-locale-and-forget-the-root-key mistake — fi/app.yml re-keyed as da: fails.

And the locale list now comes from locales.Codes(), derived from the embedded FS, so the embed directive is the single source and units_test.go follows it automatically.

Writing that test surfaced real gaps in other locales. They're in knownGaps with comments, and I've left them alone as out of scope, but three look like actual bugs rather than untranslated strings:

• fr spells three payment-term keys with hyphens — due-date, end-of-month, advance — where the templates ask for due_date, end_of_month, advanced. French payment terms render in English today.
• pl has totals.prices_include_tax where the template asks for prices_include.
• de is missing country_names.AW (Aruba); it and pl are missing KY (Cayman Islands). it and pl currency lists both stop at UYU, missing the last 14 codes.

One deviation from the exact da/nl/no mirror you verified. I added five keys those locales don't have: org.party.person, person_label, website, website_label, and billing.invoice.title.adjustment. They aren't regime-specific — the renderer uses them for any locale, and a Finnish purchase order was printing "Web: https://…" in English. It now prints "Verkkosivu:", and an adjustment credit note prints "Oikaisulasku". So fi is now 991 keys against their 986, and it's the only locale with an empty knownGaps entry alongside ar/ca/eu/gl.

🤖 Addressed by Claude Code

Merged main and pushed a26f96b — CI on the previous push caught two things, both now fixed.

The new org.party.labels.* keys from #143 are exempt, not gaps. No locale translates CUIT, NIF, RFC, P.IVA or GST Reg No. — they're country tax ID abbreviations that stay as they are in every language, so the coverage test now exempts org.party.labels.* by rule and only requires the generic labels.default. That was the test doing its job: it flagged nine new untranslated keys across all fifteen locales the moment main added them, and the right answer happened to be "these are exempt".

goconst. knownGaps is now composed from named groups (personLabels, chequePayment, currencyTail, …) so each key literal appears exactly once, which reads better than the flat table anyway. Verified with golangci-lint 2.13.1 against the repo config locally: 0 issues.

Worth flagging the maintenance shape this implies, since it's a cost the test imposes on everyone rather than just on new locales: adding an English string that locales don't yet translate will now fail CI until it's either translated, exempted, or recorded in knownGaps. That's the intent — the alternative is the silent English fallback — but if you'd rather it not block unrelated PRs, say so and I'll soften it to a report rather than a failure.

🤖 Addressed by Claude Code