Test suite's compare_numba_vs_python doesn't handle all signatures

[iguinn]

The compare_numba_vs_python function in tests/conftest.py does not handle certain function signatures correctly. One example of this behavior is that it expects the output to always have an output size (n), and doesn't handle (m) correctly. Another is that in-place functions only seem to allow one output. This seems to work for the vast majority of our processors, but there are some where it causes confusing test failures.