Handle errors in parallel teardown files

[gaborcsardi]

Currently they are ignored silently. They should probably trigger an error? (I.e. not a test failure.)

[hadley]

This is a little tricky — they should be reported as an error, but they shouldn't prevent other teardown files from running.

[gaborcsardi]

Can't we just throw the right expectation condition?

[hadley]

I spent five minutes trying to wrap the appropriate calls in test_code() (which should get errors to propagate to the reporter correctly but couldn't get it to work). I'm going to leave it for a future release.