Random tests of built distributions

The dependency hell from the last few days has taught me that things can really move under your feet. In order to catch failures like these, it might be worth pulling down random distributions and running the tests we defined in their associated recipe. If it turns out to fail, then we should raise an issue on the appropriate feedstock.

I haven't thought through the infrastructure of how to do this yet, but it may be that we want a single repository which just fetches our random distributions for each of the platforms and runs the tests.