fix(trajectory): reset state between agent runs

[linhongyu510]

Summary

Reset completion metadata when a TrajectoryRecorder starts a new run.

TraeAgent.run() reuses the agent's recorder and calls start_recording() for every execution. The method already cleared interactions and agent steps, but retained end_time, success, final_result, and execution_time from the previous finalized run. Until the second run completed, the saved trajectory could therefore report stale success and result data.

This change resets all run-scoped completion fields together with interaction history. The regression test finalizes one run, starts another with the same recorder, and verifies both in-memory and persisted JSON state.

The change is intentionally limited to run-state reset behavior; it does not alter the trajectory schema.

Validation

• pytest tests/utils/test_trajectory_recorder.py tests/agent/test_trae_agent.py -q — 8 passed
• ruff check on the changed Python files
• ruff format --check on the changed Python files
• git diff --check

AI assistance

This change was developed with AI assistance. I reviewed the recorder reuse path, split the fix from unrelated schema work, and ran the checks above.

[linhongyu510]

Friendly follow-up with a fresh verification against current origin/main (e839e55):

• branch remains 0 behind / 1 ahead
• pytest tests/utils/test_trajectory_recorder.py tests/agent/test_trae_agent.py -q — 8 passed
• Ruff check and format check passed
• git diff --check passed

The patch remains limited to resetting run-scoped completion metadata when a reused recorder starts a new run; it does not change the trajectory schema. Could a maintainer take a look when convenient? Happy to adjust anything needed.

[linhongyu510]

Revalidated on the current main (8de9be8 is 0 commits behind). Red phase: applying the regression test to the parent commit fails because the second run retains the first run's end_time, success=True, final_result, and execution_time. Green phase: 8/8 targeted trajectory and TraeAgent tests pass on this PR; Ruff format/check and git diff --check also pass. There are no competing open TrajectoryRecorder reset PRs. Maintainer review would be appreciated.Revalidated on the current main (8de9be8 is 0 commits behind). Red phase: applying the regression test to the parent commit fails because the second run retains the first run's end_time, success=True, final_result, and execution_time. Green phase: 8/8 targeted trajectory and TraeAgent tests pass on this PR; Ruff format/check and git diff --check also pass. There are no competing open TrajectoryRecorder reset PRs. Maintainer review would be appreciated.

[linhongyu510]

Revalidated on the current main (8de9be8 is 0 commits behind). Red phase: applying the regression test to the parent commit fails because the second run retains the first run's end_time, success=True, final_result, and execution_time. Green phase: 8/8 targeted trajectory and TraeAgent tests pass on this PR; Ruff format/check and git diff --check also pass. There are no competing open TrajectoryRecorder reset PRs. Maintainer review would be appreciated.