Add support for Python 3.13

[cbrnr]

In addition, bump dependencies.

[cbrnr]

NumPy errors are all in wfdb, I'll try to fix them (will link the corresponding PR once it's up).

[cbrnr]

We have to wait for MIT-LCP/wfdb-python#511 (and a new release).

[cbrnr]

This seems to work nicely. However, I don't like that some dependencies are installed in cibuildwheel.yml since everything should already be defined in pyproject.toml. Specifically, we created the cibw extras group exactly for this purpose. @larsoner do you know how the "Triage dependencies" step can be modified to use this extras group instead of manually installing packages?

[larsoner]

In theory you could modify the python -m pip install ./dist/*.whl command, probably first by getting the wheel filename then using the file:// syntax to get to something like pip install file://$WHEEL_PATH[some_new_option] and add some_new_option to pyproject.toml or whatever. TBD whether or not pip works properly with a file:// plus optional specifier, though!

[cbrnr]

I wonder why you switched to running pytest manually, since previously we used cibuildwheel functionality to run our tests (e.g., test-requires and test-command in the [tool.cibuildwheel] section, see https://cibuildwheel.pypa.io/en/stable/options/). Does this not work with your changes anymore?

[larsoner]

I wonder why you switched to running pytest manually, since previously we used cibuildwheel functionality to run our tests

FWIW it should still run tests in cibuildwheel as before. But now it does more than that.

If the cibuildwheel tests are all you had, it would only build and test on 3.9 (or 3.10 or whatever your min version is). So if you want to test multiple configs like different Python versions (e.g., 3.13, 3.11) or different dependency groupings, etc. you want a separate run. Having tests in a separate run also has the really nice side benefit of creating a clean setup to test from that doesn't have any of the stuff installed from the cibuildwheel job, so you can be more confident that things will install and run properly once deployed in the wild.

[cbrnr]

Ah, you're right, the cibuildwheel tests are still there, but they only test the wheel they've just built (i.e., using Python 3.10 currently).

It's still a bit of duplication going on because of the additional tests, which is not ideal. I'll try and see if dropping cibuildwheel (and instead building and testing wheels on runners using uv) simplifies things in a future PR.