trio_test_case: Merge internal version

[geofft]

This is mostly for discussion - this is our internal version of trio_test_case with various bells and whistles. I've lightly merged it to keep doctrings. The changes from the internal version are

• don't configure logging automatically
• unconditionally log around asyncSetUp/asyncTearDown (at INFO level, which lines up well with leaving logging unconfigured by default) instead of making it conditional on running in our CI

[catern]

See https://github.com/catern/rsyscall/commits/trio-test-case for my revised merge of this. The main thing is simplifying the warnings throwing thing - I think now it's in a nicely upstreamable state (i.e. to unittest.TestCase) while still achieving everything that the colored log messages achieved.

• I wonder if the SUSPEND_ON_FAILURE behavior is better as part of the test runner? pytest has --pdb to do this; the issue is that pdb isn't an async REPL, but AIUI no-one is using that behavior of wish in SUSPEND_ON_FAILURE right now anyway...

Also, the top answer here is amusing... https://stackoverflow.com/questions/4398967/python-unit-testing-automatically-running-the-debugger-when-a-test-fails

• Seems like logger.debug would be more appropriate than logger.info?
• Maybe there's a better way to do install_handler these days?

I'd be happy pushing this now, or waiting if we want to look at it more.