Fix orphaned simulator processes on Ctrl+C or task failure

[Jjateen]

When a task failed early (keep_going=False) or the user pressed Ctrl+C, any sail_riscv_sim processes already launched by ThreadPoolExecutor worker threads were left alive. Workers blocked in proc.communicate() kept the executor alive and the simulator ran indefinitely.

Three fixes in build.py:

• Switch SubprocessAction execution from subprocess.run() to Popen with start_new_session=True so each simulator gets its own process group, allowing os.killpg() to reach any children it spawns.
• Add _ProcTracker, a thread-safe registry of running process group IDs. Workers register their pgid on launch and deregister on completion. A shutting_down flag ensures late-registrants (processes that start after kill_all() is called) are killed immediately on add().
• Call tracker.kill_all() on early task failure (before executor shutdown) and in a try/finally so cleanup runs on Ctrl+C or any other exception. A SIGINT handler also calls kill_all() to terminate simulators before ThreadPoolExecutor waits for worker threads to join.

Fixes #1238.

[Copilot]

Pull request overview

This PR updates ACT’s Python DAG build executor to prevent orphaned sail_riscv_sim (and child) processes when the run is interrupted (Ctrl+C) or stops early due to a task failure with keep_going=False.

Changes:

• Run SubprocessAction commands via subprocess.Popen(..., start_new_session=True) and track process-group IDs for group termination.
• Add a thread-safe _ProcTracker to register/deregister running subprocess PGIDs and kill them on shutdown.
• Ensure cleanup runs on early failure and via a SIGINT handler / try/finally kill path.

[Jjateen]

@jordancarlin can you review/approve this further?