[1.0] Preserve terminal run outcomes across cleanup and failure paths

Collector cleanup can overwrite an error with completed. Failures during context construction or persistence also need the same truthful outcome contract as engine failures.

Scope

• Use the existing task/run identity and records to define consistent running, successful, failed, cancellation-requested and confirmed-cancelled outcomes.
• Ensure cleanup cannot overwrite failure or confirmed cancellation with success.
• Bring context construction, engine execution, output persistence and consumption persistence into the same outcome handling; keep run_until_idle as an idle wait rather than an implicit success result.

Acceptance criteria

• Controlled success, exception and cooperative cancellation retain their correct states after all cleanup hooks.
• Failures before engine execution and during output/consumption persistence are associated with the run and do not report success.
• Python, collector events and API-visible records agree; interruption recovery is handled by the dedicated history ticket.

Boundaries

No automatic retry/DLQ or complete workflow lifecycle implementation.

References

• Implementation dependencies: #428.
• src/flock/core/agent.py
• src/flock/core/orchestrator.py
• src/flock/api/collector.py

Start with the referenced flow in src/flock/core/agent.py and src/flock/core/orchestrator.py, then trace how src/flock/api/collector.py records task/run outcomes. Exercise success, exceptions, cooperative cancellation, context-construction failures, and persistence failures. Done means cleanup preserves terminal outcomes and Python, collector events, and API-visible records agree without making run_until_idle imply success.