feat: add pypi to conda mapping

[nichmor]

The PyPI - Conda name mapping that maps a PyPI package to it's known Conda counterpart.

This only works for packages that are available on the conda channels that it references.

This is available at https://conda-mapping.prefix.dev/pypi-to-conda-v0/{channel}/{pypi-normalized-name}.json where the channel is the name of the channel and the {pypi-normalized-name} is the normalized name of the package on PyPI.

Example:
for requests we can use https://conda-mapping.prefix.dev/pypi-to-conda-v0/conda-forge/requests.json, which will give you the corresponding json.

The schema will be saved at this location:
https://conda-mapping.prefix.dev/pypi-to-conda-v0/conda-forge/schema/schema.json

[maresb]

Thanks a lot for this @nichmor!

It seems to me like this would be a lot simpler (in my mind at least) as a table instead of a mapping, as per #11 (comment).

So for PyPI→conda-forge requests you have a list of conda-forge archives that contain a PyPI package named requests. And that list can contain some other useful columns.

For the conda-forge→PyPI requests you have the same list but instead filtered to have PyPI packages that are defined by an archive from conda-forge's requests.

This way you only need one schema and it serves both mappings. Also, when you parse published conda-forge packages, you simply need to add rows to the master table. What do you think?

[maresb]

Oh wow, thanks soooooo much everyone for taking into account my feedback, it really means a lot to me!!! 🙏 I believe that this will make it so much easier to use universally.

[tdejager]

So this PR expanded by a lot, because of the functionality I needed to test things. Based on the previous work by @nichmor, @wolfv and the suggestions by @maresb .

I did a lot of work on this branch together with CLAUDE. And basically setup a new pipeline to do the conversion, also added a small explorer to have a rich cli to explore the mappings. We also modified the README to include all of this new information. The pipelines have also been merged into a templated one, I've triggered this in CI so that the PyPI mappings are now actually online behind the v1 url.

Next up, we made a full integration test with an in-memory s3, so that we can verify that the pipelines keep working.

Finally, we added a docker-compose.yml with a minio backend so that you can recreate the mappings locally.

When all this is merged I want to add a small marimo example for @maresb that he can use to explore the RelationTable as well.