Split the test errors and runners errors

[richtja]

Right now, when the avocado runner will crash, the test itself is marked with error and the StdErrMessage will save the traceback in the test stderr. The problem is that the same will happen when the test itself will crash. Because of that, it might be difficult for users to detect if the problem is inside the test or inside the runner.

Let's make MessagesFinished and StdErrMessage only test related. And create mechanism for detecting runner specific errors.