Improve quality_eval benchmark performance and add fail-fast guardrails

Running uv run quality_eval.py often takes a very long time to complete (commonly around 56 to 83 minutes) and can appear to hang indefinitely.

Current behavior increases benchmark turnaround time and makes it hard to distinguish a slow healthy run from a stalled one.

Observed behavior
Recent completed quality runs for hosted-agent framework-agent showed durations of approximately:

56m 27s
79m 25s
82m 35s
2m 05s (fast-failure outlier)
Practical runtime for full runs appears to be around 60 to 80 minutes.

Start with quality_eval.py and run it via `uv run quality_eval.py` to trace where the 56–83 minute runtime accumulates and where a run can appear stalled. Compare the reported completed-run durations and the fast-failure case, then define guardrails that distinguish slow healthy runs from stalled ones and reduce benchmark turnaround time.

Written by the indexing model from the issue text.