Open
Description
Now that we have made a significant investment in making console output more legible, we should also test that it remains as we expect. However our console system delegates to the built-in one, which is platform dependent. The best place to stand seems to be to run the test in a separate process from the one gathering it textual output and checking it against a golden.
Likewise, #555 is about how Ava test failures look. This clearly needs a layer of wrapping so the outer Ava test succeeds only when the inner Ava test fails in the way we want. The same separate-process technique would cope with that as well.