fix(asap-tools): keep ProcessMonitor samples when a monitored process dies

[milindsrivastava1997]

Closes #740

Problem

If any monitored process exited mid-run, psutil.NoSuchProcess escaped the sampler loop, the sampler never sent its data, and remote_monitor.py skipped writing monitor_output.json — losing samples for every process. The run also continued unnoticed (the query client swallows per-query errors and runs its full schedule).

Changes

• process_monitor.py
  • Seed pid exits → roll back the partial sampling round (keeps all series equal-length), set exited_at_sample: N on the dead pid's entry, and send the data.
  • Child pid exits → stop sampling only that child; monitor keeps running.
• remote_monitor.py
  • Container client mode: wait loop checks control_pipe.poll(0) and stops the client early when the sampler reports an exit.
  • timed mode: time.sleep → control_pipe.poll(time_to_run) so it also ends early.
  • Logs an error when writing a flagged monitor_output.json.
  • Bare-metal client blocks in start(), so it doesn't stop early, but data is now preserved.
• compare_costs.py: modes with a flagged pid are skipped with an error; --allow_partial_runs includes them with a warning.
• Flag key lives in constants.PROCESS_MONITOR_EXITED_AT_SAMPLE_KEY (per-pid field, so existing consumers that iterate keys as pids are unaffected).

Testing

• New tests/test_process_monitor.py: seed death (real processes; checks alignment, rollback, and flag) and child death (mocked; monitor keeps sampling). Both fail on the previous code.
• pytest tests in asap-tools/experiments: 18 passed.
• Not yet exercised in a full e2e experiment.

🤖 Generated with Claude Code