Add Scheduled Background Job Runner

[Ezeh20]

Closes #152

Summary

Queue drains were fired from the request that enqueued them, so retries only happened if unrelated traffic came in. This adds one scheduler process that owns every recurring drain on a timer.

• src/workers/scheduler.worker.ts. Entrypoint, configurable interval
• Registers the email, notification, webhook, stellar-funding, data-export and account-lifecycle queues
• Each tick takes a lease via job-lease.service.ts, so replicas can't double-process
• Removed the fire-and-forget processQueue() calls; enqueue is unchanged
• Replaced the setInterval in server.ts; SCHEDULER_IN_PROCESS=true runs it in-process instead
• Per-queue depth/due/lag/attempt/failure metrics, and leases are released on SIGTERM
• Added the scheduler service to compose and the production image

Evidence

pnpm scheduler:verify

SCENARIO 1 — idle instance
API on :5000   : not running — no HTTP traffic is possible
before         : status=pending  attemptCount=1  error=previous attempt failed
after          : status=sent  attemptCount=2
after SIGTERM  : 6/6 leases released
RESULT         : PASS

SCENARIO 2 — two replicas, 40 due rows
attemptCounts  : attemptCount=1 -> 40 rows
lease races    : replica-A skipped 4, replica-B skipped 2
RESULT         : PASS — 40/40 processed exactly once, 0 duplicates

not