Add statsmodels support for milopy

[Zethson]

Our current milo implementation uses edgeR from R to make a few calculations here: https://github.com/theislab/pertpy/blob/development/pertpy/tools/_milo.py#L310 . However, we want pertpy to be a pure Python experience. Hence, we want to offer an alternative that does NOT use edgeR but instead uses statsmodels to do a very similar calculations.

• [ ] Figure out what exactly edgeR does here.
• [ ] Implement these steps with statsmodels.
• [ ] Compare the results with the original edgeR implementation. If we're not crazily off submit a PR and add it in another if-case

[Zethson]

CC @emdann

[emdann]

sounds good to try, although I am not sure how comparable results would be, and it would need custom implementation for some edgeR innards e.g. TMM normalization