Currently once a pytest fails, it raises an TestFailure exception with arbitrary message.  Then goofy takes the whole traceback information of that exception as the `error_msg` field of the test state.  This causes a problem that other parts of the factory software can't make decision based on the result of another test easily because `error_msg` is hard to parse.  For example, we might want to run the keyboard test which only tests the malfunction keys detected in previous round.