Harden detached-run daemon startup, submission failures, and optimizer child-pipeline loading

[Copilot]

This addresses the unresolved PR review comments on detached runs, graph optimization, and the async verification script. The changes tighten failure handling around daemon startup/submission, prevent optimizer-edited pipeline loads from hanging indefinitely, and make the verification flow self-contained.

• Detached daemon lifecycle

  • Add a startup lock around daemon creation/metadata updates so concurrent agentflow run -d invocations do not race each other.
  • Write daemon metadata atomically to avoid partially written state.
  • Emit a clear timeout diagnostic from _wait_for_daemon() with the target base URL and next-step guidance.

• Detached submission error handling

  • Catch transport failures when posting detached runs to the daemon.
  • Handle invalid/non-JSON daemon responses explicitly instead of surfacing raw exceptions.
  • Include the daemon base URL in user-facing errors for faster diagnosis.

• Graph optimizer child-pipeline loading

  • Add a conservative timeout when executing optimizer-edited pipeline.py.
  • Convert timeouts into actionable validation errors that include captured stdout/stderr when available.

• Async verification script

  • Set an explicit local git identity for the temporary seed repo commit so the verification script does not depend on global git config.

• Focused coverage

  • Add targeted tests for daemon timeout messaging, startup locking, detached submission failure paths, optimizer child-pipeline timeouts, and the verification script’s git identity setup.

try:
    result = subprocess.run(
        [sys.executable, str(path)],
        capture_output=True,
        text=True,
        cwd=str(path.parent),
        timeout=CHILD_PIPELINE_LOAD_TIMEOUT_SECONDS,
    )
except subprocess.TimeoutExpired as exc:
    raise ValueError(
        f"pipeline script `{path}` timed out after "
        f"{CHILD_PIPELINE_LOAD_TIMEOUT_SECONDS:.1f}s."
    ) from exc