Worker pools are orphaned when layup is stopped in the background

#523 and PR #561 cover the interactive case: Ctrl-C reaches the parent, which terminates the pool. A layup run started in the background is not covered, and this is the case that leaves processes running for days.

A shell backgrounding a job sets SIGINT to SIG_IGN, and the parent inherits it. The interrupt never arrives, so the except KeyboardInterrupt teardown never runs. Such a job is stopped with SIGTERM instead, whose default action kills the parent outright without unwinding, leaving the workers with PPID = 1.

Measured on a backgrounded parent with four workers, stopped by SIGTERM:

Six such processes were found on one machine recently, PPID = 1 and five days old, in an environment nobody was still using.

Suggested fix, in _run_pool: install a SIGTERM handler that raises, so the existing teardown runs, and restore the previous handler in a finally so it does not leak into a caller's process.

def _terminate(signum, frame):
    raise KeyboardInterrupt

_prev_term = signal.signal(signal.SIGTERM, _terminate)
try:
    ...  # existing pool block
finally:
    signal.signal(signal.SIGTERM, _prev_term)

Two caveats: signal.signal only works on the main thread, so this should be skipped when layup is driven from a worker thread; and SIGKILL cannot be handled by anything.