Improve default seeds used for each chain

[ricardoV94]

@ColCarroll mentioned that we might want to be more careful with how we seed different chains/processes. Right now this is our default:

pymc/pymc/sampling.py

Line 453 in 52a126d

random_seed = [np.random.randint(2 ** 30) for _ in range(chains)]

Whereas this seems to be the NumPy recommended strategy: https://numpy.org/doc/stable/reference/random/parallel.html#seedsequence-spawn

I am not sure if this also applies to other places where we create seeds such as in

pymc/pymc/model.py

Line 965 in bdd4d19

seed = self.rng_seeder.randint(2 ** 30, dtype=np.int64)

[larryshamalama]

Upon closer inspection, it seems like NumPy is recommending us to move to Generators instead of RandomStates: https://numpy.org/doc/stable/reference/random/legacy.html?highlight=randomstate#numpy.random.RandomState Should we go from RandomState to Generator, or would this be too much work?

I'm also mentioning this because each SeedSequence can spawn Generator objects and I'm not sure if this would break our code if we were to use their recommended strategy for seeding parallel processes.

[ricardoV94]

This has mostly been done. Further improvements depend on #5797