Revert back to np.random.* for test utils

[jamesETsmith]

General

Early on in development we steered away from np.random.* functions because best practice for truly reproducible results is to instantiation your own RNG and carry it around. In my opinion, we do not need need this level of stringent reproducibility in the test framework and the global state problems are not present when testing.

See this post on from the Numpy devs here.

This is something that should wait until after the initial release.