Worker crash/retry/poison message handling

[codihuston]

Cause: No explicit retry policy or poison task handling.
Analysis: Failed tasks can loop indefinitely or be lost without structured retries.
Acceptance Criteria:

• Retries are bounded with backoff and dead-letter handling.
• Poison tasks are quarantined with clear error reasons.
  Solution Approach:
• Add retry counters and failure states.
• Create a dead-letter queue or failure table.