fix: added BrokenProcessPool exception to cleanly end a collection

[ktstrader]

Description

Change to src/openhound/scheduler/service.py:

• Import BrokenProcessPool from concurrent.futures.process.
• Add a private _reset_executor() helper that shuts down the broken pool non-blocking and creates a fresh ProcessPoolExecutor with the same settings (max_workers=1, max_tasks_per_child=1).
• Add a dedicated except BrokenProcessPool branch to _handle_completed_job, placed before the existing generic except Exception. It reports the in-flight job as FAILED to BHE and calls _reset_executor(). The existing finally block continues to clear future/job_running.
• Wrap the executor.submit(...) call in _start_job with a try/except BrokenProcessPool. On failure it clears job_running/future, rebuilds the executor, and reports the job as FAILED to BHE so it does not remain in Running.

Two tests added to tests/test_bhe_job_scheduling.py:

• test_poll_recovers_from_broken_process_pool — sets a BrokenProcessPool exception on the future, asserts that job_running/future are cleared, the executor instance is replaced, and BHE receives FAILED with the expected message.
• test_start_job_recovers_when_submit_raises_broken_pool — monkeypatches executor.submit to raise BrokenProcessPool, asserts that both start_job and end_job were called on BHE, local state is cleared, and the executor instance is replaced.

Motivation and Context

Resolves: BED-8692