[bug] try using torch.multiprocessing instead of multiprocessing as

recommended at https://pytorch.org/docs/master/notes/multiprocessing.html

This is a shot in the dark to fix some multiprocessing issues in simulation and possibly pre-processing. The note about problems using fork with CUDA shouldn't be applicable, because the only places where we directly use mp (rather than letting torch handle it) are CPU-only AFAIK.

@fjay Please see if this fixes #137 (closed), though I have no way of reproducing the issue so it's hard to say for sure...