Fix blog dates ignoring the page locale

[cverorg]

Summary

Fixes #34.

• fmtDate(s, format, lang) gains a third lang argument: the page's own BCP-47 locale (the Lingo variant actually being rendered). Its default/unrecognised branch now formats with Intl.DateTimeFormat(toBcp47(lang), { year: 'numeric', month: 'long', day: 'numeric' }) instead of an unconditional en-US literal. When no locale is known, output stays byte-identical to the historic en-US string.
• Every fmtDate caller (ArchiveView.astro, BlogIndexView.astro, PostArticle.astro, and the /search/label/[tag] archive routes) now passes the page's own meta.lang, not the site default.
• cjk is accepted as an alias of cjk-full.
• An unrecognised blog-date-format value still renders (the locale default, never a blank date), and now emits one build-time warning naming the allowed values — once per build per bad value, not once per post.
• A site declaring a non-English lang with no blog-date-format set will see its dates change (e.g. lang: zh-Hant moves from July 13, 2024 to 2024年7月13日) — this is the point of the fix, but is a real, visible output change for any such site's existing pages.

Round 1 review fixes

• P1 — invalid lang crashed the whole build. A lang: value that is not a well-formed BCP-47 tag (ja_JP, zh_TW, en US, non-ASCII text) made Intl.DateTimeFormat throw RangeError, taking the whole build down on the first date it rendered — a regression versus main, which always rendered (in English) regardless of lang. fmtDate's locale-default renderer now tries the page's tag and falls back to the historic en-US output, byte-identical to main, on any tag Intl rejects; never throws.
• P2-1 — an explicit CJK format was applied to every locale, including English. Ruling (2026-09-11): an explicit blog-date-format is the site owner's choice for the locales it fits. cjk / cjk-full / cjk-badge / cjk-md are a script, not a punctuation style, so they now apply only when the page's own locale is Japanese, Korean, or Chinese (ja/ko/zh-*) — a non-CJK page locale renders that locale's own Intl default instead (so an en-US page stays July 13, 2024 even when the site sets blog-date-format: cjk). ymd-slash has no script and is unaffected — it always applies, on every locale, as before. When no page locale is known at all (lang absent), a named CJK format still always renders, matching the pre-existing behaviour before lang was threaded through. This rule is documented at the CJK_DATE_FORMATS/isCJKLang definitions and the fmtDate doc comment in packages/sitetile/astro/src/lib/blog.mjs.

Cheap round 1 P3s also applied

• Fixed a timezone-dependent test suite (blog-date-format.test.mjs pins TZ=UTC for its own process) — the underlying UTC-midnight-vs-local-date gap is pre-existing elsewhere and out of scope here.
• Added a hardcoded ICU literal assertion (ja-JP → 2024年7月13日) so a small-ICU/system-ICU runner missing Japanese locale data fails loudly instead of silently passing on an Intl-vs-Intl comparison.
• Replaced the structural regex in chrome-copy.test.mjs that pinned fmtDate's exact default-branch source text with a behavioural check (it would have broken on the P1 fix's try/catch for no reason related to what it was guarding).
• Added a structural guard requiring every fmtDate( call site to pass a lang argument, mirroring the existing dateBadgeParts guard.
• Added a test asserting every KNOWN_DATE_FORMATS entry renders without an "unrecognised format" warning.

Test plan

• node packages/sitetile/blog-date-format.test.mjs — 9 tests: no-format under ja-JP/zh-TW/en-US against Intl's own output plus a pinned ICU literal, no-format-no-locale byte-identical, cjk alias, the new CJK-locale-gating rule (ja-JP/zh-TW/ko-KR stay CJK, en-US falls back, ymd-slash always applies), unrecognised value warns once and still renders the locale default, every KNOWN_DATE_FORMATS value warns never, every existing explicit format unchanged, unparseable date passthrough, and the new invalid-BCP-47-tag guard (garbage tags, empty tag, non-CJK bad tag under a CJK format) never throws and matches main's byte-identical en-US output.
• Same suite re-run under TZ=America/Los_Angeles — all 9 pass.
• node packages/sitetile/chrome-copy.test.mjs — 15 tests, including the new behavioural (not structural) CJK-neutrality check and the new fmtDate call-site guard.
• node packages/sitetile/blog-locale-archives.test.mjs, blog-unlisted.test.mjs, blog-defaults.test.mjs, blog-tenancy.test.mjs, blog-url-gating.test.mjs — unaffected, all green (28/18/2/14/15).

Not verified

• No end-to-end Astro build/smoke test was run (packages/sitetile/astro/node_modules is not installed in this environment), so the actual rendered HTML on a live multi-locale build was not visually re-checked this round — only the underlying fmtDate/call-site changes were unit-tested.

[cverorg]

Round 1 fixed: fmtDate no longer crashes on an invalid BCP-47 lang tag (falls back to main's byte-identical en-US output), and an explicit CJK blog-date-format now applies only to CJK page locales (ja/ko/zh-*) rather than every locale — ymd-slash stays unaffected; details and the cheap P3s applied are in the updated PR description.