bug: Execution shows Failed then Success after cleanup (race condition)

[vybe]

Summary

Scheduled task executions briefly show "Failed" status with error "Stale Execution Slot TTL Expired for Agent [name] Cleaned by Cleanup Service", but the task actually succeeds and status updates to "Success" ~14 seconds later.

Context

Reported by Alex Sazonenka on Paradigm Live instance (2026-04-17). A generation task at 15:44:13 showed Failed immediately, but after page refresh 14 seconds later it was Success with full output and logs available.

This creates confusion and would cause false-positive failure notifications if we implement alert channels.

Root Cause Hypothesis

Race condition between:

1. Cleanup service's TTL watchdog marking execution as stale/failed
2. Actual execution completing successfully

The cleanup service is winning the race and marking the slot as expired before the execution can report completion.

Acceptance Criteria

• Cleanup service checks for actual execution completion before marking failed
• Add grace period or completion signal handshake
• Execution status is accurate on first render (no phantom failures)
• Add test case for concurrent cleanup + completion scenario

Technical Notes

Related files:

• src/backend/services/cleanup_service.py - TTL watchdog
• src/backend/services/slot_service.py - Slot management
• src/backend/services/task_execution_service.py - Execution lifecycle

May need to check execution's actual process state before cleanup marks it failed.