fix(l10n): disambiguate duration "m" label from millions abbreviation with msgctxt

[parh0m2007]

Problem

The single-character msgid "m" was shared by three different meanings:

1. Millions — small_count() in src/sentry/templatetags/sentry_helpers.py (_("m"))
2. Extra-short month label — DURATION_LABELS in static/app/utils/duration/getDuration.tsx
3. Extra-short minute label — DURATION_LABELS in static/app/utils/duration/getDuration.tsx

Because gettext maps one msgid to one msgstr per locale, the Russian catalog carries the millions translation («млн») for all three. With unitStyle="extraShort" (TimeSince → getRelativeDate → getDuration), durations under a month rendered as «5млн» ("5 millions") instead of «5м» ("5 min").

Fixes #124240

Solution

Route the extra-short month/minute labels through gettext translation contexts (msgctxt), keeping small_count()'s millions meaning as the context-less "m". Jed already supports pgettext lookups at runtime (msgctxt\u0004msgid keys), and the extraction script already knows the pgettext signature — the pipeline pieces just weren't wired up for frontend use.

Changes

• static/app/locale.tsx: add tp (a pgettext wrapper) alongside t/tn, so ambiguous single-letter msgids can be disambiguated with a context.
• static/app/utils/duration/getDuration.tsx: the extra-short month/minute labels now use tp('duration', 'm').
• build-utils/ts-extract-gettext.ts: register tp: ['msgctxt', 'msgid'] so context-bearing calls are extracted with their msgctxt.
• build-utils/po-catalog-loader.ts: previously only read translations[''], silently dropping all context-specific entries. Now iterates all contexts and emits entries in the msgctxt\u0004msgid key format Jed looks up.
• bin/merge-catalogs: catalog.get(frontend_msg.id) matched context-specific msgids against the context-less entry (the source of the collision). Now looks up with catalog.get(id, context); the context travels with the Message when written back.
• en/ru catalogs: surgically add the new msgctxt "duration" entry (ru: «м»; other locales fill in via Transifex). The next make merge-locale-catalogs + Transifex sync will refresh references across the remaining locales.

small_count() and its «млн» translation are untouched.

Verification

• pnpm run typecheck — clean
• pnpm run lint:js (oxlint) on changed files — 0 errors
• Jest: all duration/locale/timeSince/relativeTime suites pass (126 tests), including two new regression tests asserting that with a Russian-style catalog getDuration(122, 0, false, true) → '2м' and months → '3м' (not «млн»)
• End-to-end: pnpm run build-js-po extracts the msgctxt "duration" entry; bin/merge-catalogs verified to keep plain "m" → «млн» and create the separate context entry; po-catalog-loader output verified to resolve via Jed.pgettext('duration', 'm')

[github-actions]

🚨 Warning: This pull request contains Frontend and Backend changes!

It's discouraged to make changes to Sentry's Frontend and Backend in a single pull request. The Frontend and Backend are not atomically deployed. If the changes are interdependent of each other, they must be separated into two pull requests and be made forward or backwards compatible, such that the Backend or Frontend can be safely deployed independently.

Have questions? Please ask in the #discuss-dev-infra channel.

[parh0m2007]

The usr_comments → user_comments typo is fixed in e0dc82c. It pre-dated this PR (present in upstream/master), but it was a real crash path: any frontend message carrying translator comments (babel's user_comments) would raise an AttributeError in merge_message(). Verified the previously-crashing path now works and the comment merges into the catalog.