Python 3.14 forkserver does not avoid the GNU OpenMP fork abort

The usual fix for 'Terminating: fork() called from a process already using GNU OpenMP' (numba parallel=True under the omp layer, or other libgomp users) is a non-fork start method. On CPython 3.14 the default forkserver preloads __main__, so if the script calls a parallel kernel at module level, outside the if __name__ == '__main__' guard, the server process starts OpenMP and every worker forked from it aborts. multiprocessing.Pool then hangs forever, while ProcessPoolExecutor raises BrokenProcessPool. Checked with numba 0.68.1 and system libgomp: same script works with the call moved under the main guard, with spawn, or with NUMBA_THREADING_LAYER=workqueue.