Better testing of CI scripts

[jonhoo]

Currently, we just make sure that CI passes, but ideally we'd have a way to see that the CI scripts actually run the right commands in every job given the configuration options for that job. To do this, we'll probably need to run something in our own CI at the very end that walks through all of the jobs we ran and checks the arguments of every command run. Not sure how we do that, or whether Azure's API would even let us do it, but something worth considering. In theory that would catch things like #50.