Joshua should handle non utf-8 characters more gracefully

Currently Joshua assumes all test outputs are valid utf-8 characters and it crashes when seeing invalid utf-8 characters. It's desirable that this can be fixed so that Joshua could escape/ignore invalid utf-8 characters instead of crashing.