#310 Record stop_reason=failed on warmup or iter failure

[jathavaan]

Summary

• When a benchmark fails during warmup, the @monitor decorator now records stop_reason=failed regardless of use_sequential_stopping. Previously the non-sequential branch (Databricks national-scale + DuckDB/PostGIS national-scale) fell through to stop_reason=fixed even when no timed iteration ran.
• The fallback now applies only on the success path: failure paths always set failed.
• Closes stop_reason=fixed reported when warmup raises in non-sequential @monitor branch #310.

Test plan

• Trigger a warmup failure on a non-sequential benchmark (e.g. Databricks default-2-nodes-large); confirm benchmark_metadata.parquet records stop_reason=failed and achieved_iterations=0.
• Run a healthy non-sequential benchmark (e.g. Databricks broadcast-4-nodes-large); confirm stop_reason=fixed and achieved_iterations=5.
• Run a sequential benchmark to completion (PIP small); confirm stop_reason=precision still records correctly.
• Force a timed-iter failure mid-run on a sequential benchmark; confirm stop_reason=failed and partial samples persist.

[Copilot]

Pull request overview

This PR fixes incorrect stop_reason reporting in the @monitor benchmarking decorator when a benchmark fails during warmup or iteration execution—specifically ensuring non-sequential benchmarks no longer fall back to stop_reason=fixed on failure paths.

Changes:

• Ensure stop_reason is set to StopReason.FAILED whenever failure is not None, independent of use_sequential_stopping.
• Restrict the existing stop_reason fallback logic to the success path only (i.e., when no failure occurred).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.