Add function to run both the numba and pure python versions of some function

[slwatkins]

As discussed in #241 and proposed in #241 (comment), we can write a function that takes a numba-wrapped function, and automatically runs both the numba version and the unwrapped pure python version. For testing these numba-wrapped functions, it is necessary to run the pure python version to trigger the code coverage.

I've added a fixture to tests/conftest.py, which returns such a function for some generic wrapped function in, e.g., dsp. I've also updated the fixed_pickoff_time test to use this function as an example, which cleans up the test substantially.

I'll keep this as a draft for now, as I'm curious if there's a better place to store this function or if there are any thoughts on how I'm using it. I also want to point out that it doesn't work well with pytest.raises, so I'm just using inspect.unwrap for checking if the errors are being raised for both cases.

---

Before submitting a pull request, please make sure you've read and understood the pygama developer's guide: https://pygama.readthedocs.io/en/latest/developer.html. In particular, do not forget to:

• Conform to our coding conventions
• Update existing or add new tests
• Update existing or add new documentation
• Address any issue reported by GitHub checks

[gipert]

Like! Can be merged from my point of view, after adding a docstring and a note to the dev's guide. I am curious to see whether vectorization will yield results which differ from the pure python implementation by more than machine's precision.

[slwatkins]

@gipert I added some documentation on using it, and I think it should be ready to go - let me know if there's any follow up comments on what I've changed!

[jasondet]

This is awesome @slwatkins thanks! I'll merge it once the checks are complete