crosscat misappropriates entropy from numpy

[riastradh-probcomp]

simple_predictive_sample_unobserved calls numpy.random.multinomial with no specified seed, so numpy presumably draws it from the environment.

[fsaad]

This is probably an issue, but notice that simple_predictive_sample when invoked through simple_predictive_sample_multistate, does have the numpy seed managed
https://github.com/probcomp/crosscat/blob/master/crosscat/utils/sample_utils.py#L257
The problem is when we invoke simple_predictive_sample externally -- which only happens when the
crosscat instance has 1 model (which is rarely the case, and why I think the effects of the bug have not been witnessed).

[riastradh-probcomp]

Unrelated. That random state is correctly deterministically derived from the user-specified seed; the one ine simple_predictive_sample_*unobserved* is not.

[axch]

Fixed by e5bab7a .