Manage numpy random state with a pyttb specific generator

[ntjohnson1]

Right now we just have raw calls to np.random. However, if we want to seed things we need to do it globally for numpy which isn't recommended. We should come up with a procedure to have a scoped (or multiple scoped generators) through pyttb.

I can image two approaches (but there is some complexity for user provided methods that generate random numbers).

1. Create a pyttb singleton wrapper that a user can set then we globally use that.
2. Plumb through random generator everywhere to avoid the singleton pattern