Use np.RansdomState in tests - #1312

I've restarted the build to hopefully fix the other test that was randomly broken in master. Once that goes though, I'm 💯 on this getting merged.

@nden - there is still a failure - is there a test for which the random state is still not being set?

@nden - since the test is still failing but should be deterministic, maybe this is just a case of changing the random seed?

@astrofrog This was my thought too but shouldn't this be also deterministic across machines. The tests pass for me locally. I changed it we'll see what happens now. (Sorry for not being very responsive. I have to finish something else before I leave for a month at the end of August. I'll also try to do as much as I can for the 0.3 release but it may not be much.)

The tests pass, but I'm going to restart Travis to double check.

@astrofrog Tests passed again. There's nothing controversial about this so I am going to merge it. If we see this test failing again we'll revisit

@nden - sounds good!

@nden - this is perfectly fine as it was merged, but I just wanted to point out astropy.utils.misc.NumpyRNGContext - you can see an example of its use at https://github.com/astropy/astropy/blob/master/astropy/stats/tests/test_funcs.py . It's basically a decorator that sets the numpy RNG seed so you can just use the regular np.random.* functions but have them be deterministic.

RandomState should do essentially the same thing, so whichever is easier in a given case is fine.

@eteq Thanks for pointing this out, I didn't know about it.