feat: implement Phase 2 — worker pool, backoff strategies, and retry fix

[reckziegelwilliam]

Core changes:

• Add internal/worker package: Handler registry, BackoffStrategy interface, FixedBackoff, ExponentialBackoff, Worker (single consumer), Pool (multi-queue multi-worker orchestrator) with graceful shutdown and per-worker state snapshots
• Fix retry bug in both backends: Nack previously set non-final failures to "failed" status, which Reserve never picks up (it only selects "pending" jobs). Non-final failures now go back to "pending" with scheduled_at = now + retryDelay
• Extend Backend.Nack signature to accept retryDelay time.Duration so the worker can pass its computed backoff to the backend atomically
• Update all backend tests to use the new Nack signature and assert "pending" status (not "failed") for retryable jobs
• Mark Phase 2 complete in PLAN.md