Make this work on Apple laptop

[jperla]

1. Ignore pycache
2. multiprocessing is crashing, so default to 0, but make it easy to switch to 2 or more workers if needed
3. Conda uses an environment.yml not requirements.txt (pip), so ensure we use that and take out packages which were not available on Apple M2, which don't appear to be necessary
4. Use MPS graphics card on M2 (cuda is for Nvidia). This speeds up the NN 10X.