Look into jupytext for managing doc source notebooks

[mwaskom]

Would be nice to be able to edit prose, etc. in plain text, along with not having to track all the cell metadata, etc.

[mwaskom]

Looked into this a little bit.

The main advantage of jupytext is that a plain, less-structured (than json) text document can be the authoritative source for a file that you (also) interact with as a notebook. That document will produce cleaner diffs in the git repo. And for the doc source, where edits are often editing/fixing of the expository text, it's more convenient to open a plain text file right in vs code than to find the notebook in jupyterlab, edit, and clear outputs.

It seems there would be two possible models

1. Have plain-text source files in MyST format as what is tracked in git. Then the first step of the doc build would be to "sync" all the tracked files to produce .ipynbs, and the existing doc build chain would proceed from there.

2. Have plain-text source files in MyST, use jupytext so that they can be edited in a notebook interface (useful for iterating on the plots), but then rework the doc build to go straight from the MyST files into sphinx. This is more work but also potentially would produce a more robust doc infrastructure. It's also possible that it wouldn't support all of the weird parts of the seaborn doc build (e.g including API example notebooks in the docstrings).

[mwaskom]

Looked into this a little further for v0.12 and think we're gonna stick with ipynb for now, for better or worse.