Bug: PID file is written before signal handlers are registered, leaving a window where SIGTERM is unhandled

[ChaseFlorell]

Problem

In do_start, _prepare_stream_process (which calls write_pid_file) is called before register_signal_handlers(). If a SIGTERM or SIGINT arrives in the window between those two lines, the signal is handled by the default OS handler: the process dies immediately without:

• Writing the stop sentinel (stream.stop)
• Cleaning up the PID file
• Transitioning the broadcast to complete

The result is an orphaned broadcast in a non-complete state and a stale PID file that will cause the next --start to attempt to kill a process that no longer exists.

Proposed Fix

Register signal handlers before writing the PID file:

register_signal_handlers(config, logger)
write_pid_file()

Acceptance Criteria

• register_signal_handlers is called before write_pid_file in do_start
• A SIGTERM received immediately after --start begins still triggers graceful cleanup
• The PID file and stop sentinel are cleaned up on any termination path

Proposed Tests

• test_signal_handlers_registered_before_pid_write — use unittest.mock to assert register_signal_handlers is called before write_pid_file within do_start
• test_sigterm_during_startup_cleans_up — mock signal delivery immediately after handler registration; assert PID file is removed and stop sentinel is not left behind