Reseed RNG before each test

[casperisfine]

This ensure that for a given --seed value, a test will always start from the same seed.

Otherwise previous tests using the RNG, would leak state.

This is something we had for a very long time as a global setup, I figured it made sense upstream.

A recent commit to ruby reminded me of it: ruby/ruby@7c1553e

[zenspider]

I dunno about this one. I suspect it'll break a lot of tests out there.

I have sranded before shuffling classes and methods for each runnable so that will stabilize that aspect (previously user/test use of srand could throw run order off)...

without more analysis to show that this won't "break the world" I think I'd rather leave it up to the user/test to do it themselves. I've added Minitest.seed in order to make this more readily available.