feat: Implement automatic reclaimer backoff jitter to prevent thundering herd #122

Feature Summary

The queue reclaimer processes pending or stalled submissions across the Redis 3-tier priority streams (ac:submissions:run, submit, bulk). Under high node restart events or cluster network recovery, concurrent reclaimers may query Redis at the same millisecond, risking a thundering herd.

Proposed Implementation

1. Add full jitter (rand.Float64() * baseInterval) to the reclaimer loop sleep interval in internal/queue.
2. Add Prometheus counter ac_reclaimer_claims_total{stream="...", result="recovered|poison"}.
3. Provide unit test verifying jitter distribution and absence of synchronized bursts.

Acceptance Criteria

• Jittered interval prevents synchronous lock-step queries.
• Clean metrics incrementing upon each claim cycle.