feat: adding a very basic FSSpecSource

[lobis]

This PR simply adds the changes from #692 with the necessary updates to make it compatible with current uproot.

A very basic Source is implemented for fsspec allowing the user to optionally use fsspec by setting the appropiate handler in the read operation.

with uproot.open(
    "https://github.com/CoffeaTeam/coffea/raw/master/tests/samples/nano_dy.root",
    http_handler=uproot.source.fsspec.FSSpecSource,
) as f:
    data = f["Events/MET_pt"].array(library="np")
    assert len(data) == 40

[jpivarski]

Great to see this starting! (We'll talk on Zoom in an hour.) I'm watching it, but I removed myself as a reviewer for now because you can use the "request review" button as a way of reminding me about it when it's done. I don't think I get emails about a PR going out of draft mode, but I do get emails about requests-to-review, so we use that as a last step in our process.

[lobis]

@jpivarski @nsmith- I think this is ready to go.

Currently the fsspec source works in place of the http, file, s3 and xrootd sources. At this point fsspec is only present as a test dependency and this can be merged without causing any problems as it's fully optional.

One of my next steps (for a dedicated PR) would be to refactor the reading.open options for handlers (http_handler, file_handler, etc.) in order to use a single handler option (handler) that can be set to the currenty existing sources or the new fsspec source with the goal of setting the fsspec source as the default once the integration is complete.

I think this would introduce some limitations such as the ability to open some files with one protocol and some with another in the same function call, but as we discussed this is an acceptable compromise. The ability to use the previous sources (http, s3, etc.) will remain and the corresponding source will be assigned from parsing the uri string.

The CI starting to fail when I added s3fs as a dependency (required by fsspec to read s3 files. I have no idea why, it looks like it has something to do with ssl and only fails the CI on some python versions, not all. Any idea what might be going on? (https://github.com/scikit-hep/uproot5/actions/runs/6408524082 for the logs).

I saw that minio is used to add support for s3 in the current implementation of the s3 source, currently being an optional dependency. When this project is done fsspec will become a core dependency and users trying to open s3 files will recieve an error (from fsspec) that will direct them to install s3fs. s3fs should still be present on the test dependencies to validate against s3 files though.

[jpivarski]

This looks great! It adds a new class (which doesn't disturb any existing code yet) and it's following all of our local conventions. I think it's ready to merge.

[jpivarski]

It's a blocking call in this first implementatation, but will be made asynchronous later.