refactor: hiv

[tristan-f-r]

Adds documentation back to the HIV dataset, and:

• Does offline UniProt mapping. This drops a significant part of the code.
• Drops KEGG gold standard generation, since it wasn't sufficient. Note that we now include a Prior work section in datasets/README.md, so this will never be actually lost.

[agitter]

I left some initial comments that mostly pertain to my different expectations for what can go in the dataset readmes. I am finding it hard to review the new script-based pipeline with respect to the original notebooks. I'm not sure if it is a pure port or a complete rewrite.

[tristan-f-r]

The kegg_orthology.py is a port. Every other file has been substantially rewritten, especially name_mapping.py, though for the better: this entire pipeline is now just 90 lines of Python.

[tristan-f-r]

I'm going to separate the miscellaneous cache changes over to make this an easier diff to read.