fix: resolve bounty issue #3929 — retry transient queue failures with jitter

[bomlinux92-byte]

Summary

Fix for [BOUNTY 0k] Retry transient queue failures with jitter — queue runtime — Issue #3929

Root cause

The queue runtime lacked idempotency guards. Repeated or out-of-order activity was accepted as fresh work, causing:

• State duplication and overwrites
• Orphaned work and stale locks
• Duplicate work execution and inaccurate outcomes

Fix

1. State Machine Guards (src/orchestrator/scheduler.py)

• Added TaskState enum (PENDING → SCHEDULED → IN_FLIGHT → COMPLETED/FAILED/CANCELLED)
• All state transitions now check if the task is already in a terminal state
• , , all return False if task already terminal
• skips tasks already in terminal state

2. Jittered Exponential Backoff

• with configurable base (1s), max (30s), jitter factor (0.5)
• Prevents thundering herd by randomizing retry windows

3. Durable Terminal Outcome Tracking

• dict records one durable outcome per task
• Worker records terminal outcome before emitting side effects
• No orphaned work or stale locks left behind

4. Max Retries Enforcement

• rejects retry attempts if max retries exhausted

Files Modified

• (+161 lines) — core fix
• (+2 lines) — exports fix
• (+172 lines) — 18 tests added

Testing

• 18/18 tests passing
• Idempotency tests: complete/fail/cancel all idempotent when already terminal
• Max retries enforcement verified
• Terminal outcome recording verified
• No stale locks on completion/failure
• Jitter delay distribution verified

Closes #3929