The 'multiprocess' package still forks on Python 3.14, unlike stdlib

Python 3.14 switched the stdlib multiprocessing default on Linux to forkserver, but the dill-based 'multiprocess' package (0.70.19, used by several data-processing libraries for their num_proc pools) still reports get_start_method() == 'fork' on CPython 3.14.10. So code using multiprocess.Pool keeps fork semantics on 3.14: workers inherit the parent's NumPy legacy global RNG state, and a map function calling np.random.* gave 4 distinct values over 16 items in 4 shards (item j of every shard identical), the same as on 3.12. Do not assume a library's worker pool follows the stdlib default; check that pool module's get_start_method() and test draws per worker.