Backend: Cache localized GetLanguages/GetRegions responses per locale

[aapeliv]

Since #9173, Resources/GetLanguages localizes all ~300 language names on every request. Each name goes through babel.Locale.parse, and ~120 of our ISO639-3 codes aren't known to babel at all — for those, babel does an uncached filesystem stat and raises UnknownLocaleError each time. That adds up to ~38ms of pure CPU per request (measured warm), on an endpoint that used to be a cached-dict lookup — which is what caused the backend CPU usage spike starting 2026-06-28.

The responses are fully static per locale, so this caches the built response protos in an lru_cache keyed on the babel locale (bounded by the set of supported UI locales). Per-request cost returns to effectively zero; the first request per locale after startup pays the localization cost once. GetRegions gets the same treatment (it was cheap, but it's the same shape).

Sharing one proto message across requests is safe here: nothing mutates responses after the servicer returns them — the tracing interceptor's sensitive-field sanitization deepcopies first — and there's a comment in the servicer noting this caveat.

Also removes the now-unused per-code try_localize_* wrappers from LocalizationContext, since the servicer now calls the module-level functions from couchers/i18n/localize.py directly (those remain and are still tested).

Testing

• Benchmarked the old path warm: ~38ms CPU per GetLanguages call (~24ms of that in the exception path for the 122 codes babel doesn't know); the new path is an lru_cache hit after the first call per locale.
• uv run pytest src/tests/test_resources.py src/tests/test_localize.py — 13 passed. test_resources.py already covers localized language/region names for non-English locales, which exercises the cached path.
• make format and make mypy pass.

Backend checklist

• Added tests for any new code or added a regression test if fixing a bug (covered by existing localization tests in test_resources.py/test_localize.py)
• Run the backend locally and it works
• Added migrations if there are any database changes, rebased onto develop if necessary for linear migration history (no database changes)

For maintainers

• Maintainers can push commits to my branch
• Maintainers can merge this PR for me

This PR was created with the Couchers PR skill.

[CouchersBot]

Mobile

Install the Dev Tool (iOS via TestFlight, iOS Simulator, or Android .apk) here.

Scan the QR with your phone camera, or tap Open in Dev Tool on the device, to open this branch in the installed Dev Tool dev client.

iOS	Android
Open in Dev Tool	Open in Dev Tool

Deep links

iOS

couchers-devtool://expo-development-client/?url=https%3A%2F%2Fc5984238--ota.preview.couchershq.org%2Fios%2Fmanifest

Android

couchers-devtool://expo-development-client/?url=https%3A%2F%2Fc5984238--ota.preview.couchershq.org%2Fandroid%2Fmanifest

Web (Vercel)

View the Vercel web preview for this branch.

Backend

Schema	Schema diff	Sample emails

Other

Protos	Backend coverage	Web coverage

[tristanlabelle]

Thanks!

[github-actions]

📝 Release Notes

This PR does not need to be included in release notes.

Reason: This is a backend performance optimization that fixes CPU usage on resource endpoints, but it does not introduce a user-visible feature or a noticeable end-user change on its own. While important operationally, it is closer to internal infrastructure/performance maintenance than a major platform-facing improvement users would recognize in release notes.

🤖 Bot Debug Information

Model: couchers.openai.gpt-5.4
Decision: exclude
Reasoning: This is a backend performance optimization that fixes CPU usage on resource endpoints, but it does not introduce a user-visible feature or a noticeable end-user change on its own. While important operationally, it is closer to internal infrastructure/performance maintenance than a major platform-facing improvement users would recognize in release notes.