Sync Scheduler job trigger engine

Implement as job-trigger-engine.ts with a single exported async function runScheduledJobs(supabaseClient, integrationEdgeFunctionUrl). Use a p-limit or manual concurrency pool (not raw Promise.all on large arrays) to cap parallelism at 5. For duplicate prevention, use Postgres advisory locks: SELECT pg_try_advisory_xact_lock(hashtext(org_id || integration_id)) inside a transaction before inserting the pending row. If the lock is not acquired, skip.

Status transitions: use UPDATE ... WHERE status = 'pending' RETURNING * to detect race conditions — if 0 rows updated, another instance won the race. For the structured summary, collect results from all Promise.allSettled results and build the summary before returning. Handle the case where the cron config reader returns 0 due integrations gracefully — return immediately with triggered: 0 without error.

Unit tests with mocked Supabase client and Integration Edge Function: (1) single due integration → inserts pending, updates to running, updates to completed; (2) Integration EF returns error → updates to failed with error captured; (3) already-running integration (within 30 min) → skipped, not duplicate-triggered; (4) two concurrent engine calls → advisory lock prevents duplicate runs; (5) parallel execution — 5 integrations triggered concurrently, all complete. Integration test: full end-to-end with Supabase local dev — seed organization_integrations with 3 due entries, invoke scheduler, assert sync_run_log contains 3 completed rows. Load test: 20 due integrations complete within 50s.