Lazy-load translations and batch selection deletion bookkeeping

[frankrousseau]

Problems

• All 15 translated locales were bundled eagerly into the main chunk: ~1.5 MB minified shipped to every user for languages they never open, putting the entry bundle at ~2.3 MB.
• deleteSelectedShots dispatched one deleteShot per selected shot, and each commit paid three full list passes (cache.shots, cache.result, displayedShots) — deleting N shots from a list of M cost O(N×M) of store work on top of the N HTTP calls. deleteSelectedAssets had the identical pattern. A failure mid-loop also left no trace of what had already been deleted.

Solutions

• Locales now load through dynamic imports — one chunk per language (~100 kB each), registered via setLocaleMessage when setLocale asks for it. The main chunk drops from ~2.3 MB to 763 kB. English and its production-style overlays (en_nft, en_video-game) stay eager since en is the fallback locale and App.vue switches to the overlays synchronously. setLocale keeps its signature but returns a promise; a pending-language guard keeps a slow older chunk from overriding a newer switch.
• Deletion requests stay serial — no parallel hammering of the server — but the store bookkeeping accumulates and lands in a single bulk mutation (REMOVE_SHOTS / REMOVE_ASSETS): one filter pass per list and one stats recompute, whatever the selection size. A finally block commits what the server actually deleted even when a request fails mid-loop, so the UI never desyncs on partial failure. Two deliberate improvements over the sequential path: the displayed count is recomputed after pure removals (it previously went stale until the next search), and canceled flags are set before the assets stats recompute so canceled assets stop counting immediately.

Tests: bulk/sequential equivalence pinned for shots, serial requests asserted (maxInFlight === 1), single-commit and partial-failure paths covered for both stores; lang spec extended with the lazy-load and rapid-switch cases. Full suite: 88 files, 1038 tests, 0 failures.