Normalize runner status handling and ensure timeout cleanup in execution API

[fractional-pm]

Motivation

• Ensure the execution endpoint correctly interprets varied runner status strings and persists accurate execution status.
• Prevent the request timeout from leaking by guaranteeing clearTimeout is always called.
• Allow the API to represent an in-progress state (RUNNING) separately from success and error.

Description

• Added normalizeRunnerStatus to map various runner status strings (e.g. COMPLETED, DONE, FAILED) to a canonical SUCCESS/ERROR/RUNNING set.
• Broadened the local status variable to include RUNNING and preserve the initial execution record as RUNNING when created.
• Wrapped the runner fetch call in a try/finally so the timeout is always cleared, and set status based on the normalized runner response payload.
• Kept the existing error handling to mark executions as ERROR on exceptions and only set completedAt when status === 'SUCCESS'.

Testing

• Ran the automated test suite via pnpm test and all tests completed successfully.
• Ran a type check/build via pnpm build to validate typings and the build and it succeeded.

---

Codex Task