Random number generation

[fdmalone]

We currently do a bit of a hack to seed the RNGs when running in parallel (a random seed is chosen and other processors are seeded using an offset), which probably isn't the most robust thing to do.

numpy now has better support for this type of scenario: https://numpy.org/doc/stable/reference/random/parallel.html (xref https://scicomp.stackexchange.com/questions/23547/parallel-mersenne-twister-for-monte-carlo) and we should probably do that instead with some support for backwards reproducibility (which would not be the default but could still work).