refactor: optimize cron scheduling, DB maintenance, and TMDB reliability

- Skip recommendations refresh for titles fetched within the last 7 days via `getTitlesWithFreshRecommendations`; only stale titles are re-fetched
- Replace `Promise.allSettled` in `refreshTvChildren` with a `mapWithConcurrency` helper (concurrency = 5) to stay within TMDB's rate limit
- Add 30s request timeout middleware to TMDB client to prevent indefinitely hung fetches
- Expand `optimizeDb` weekly job to prune 30-day-old cron runs, integration events, expired sessions, and expired verifications
- Add `wal_checkpoint(TRUNCATE)` to `optimizeDatabase` so the WAL is flushed before `PRAGMA optimize`
- Add `recoverStaleImportJobs` called on server startup to mark any `pending`/`running` import jobs as errored after a crash/restart
- Return `false` from `triggerJob` immediately if the job is already running

docs/vercel.json

@@ -0,0 +1,3 @@
+{
+  "bunVersion": "1.x"
+}