REL: upgrade cuilbuildwheel (2.3.0 -> 2.4.0)

[neutrinoceros]

PR Summary

Simple one line patch so we can use the current version of cibuildwheel for our next release

the one patch that I think is relevant to us is: pypa/cibuildwheel#975

see complete release notes : https://github.com/pypa/cibuildwheel/releases

A test job is running on my fork for this branch: https://github.com/neutrinoceros/yt/actions/runs/2140271185

[matthewturk]

Looks good to me. If it passes on your branch, merge it.

[neutrinoceros]

actually it doesn't, I'm reporting a regression upstream.

[matthewturk]

Well then! ;-)

[neutrinoceros]

Here's the report pypa/cibuildwheel#1078
This upgrade isn't necessary for the release, as long as the currently pinned version continues to work, so let's just close this for now. I'll just reopen in case the problem can be adressed from our side or if a new, fixed version of cibuildwheels lands before we need it.

[henryiii]

Are you sure this isn't a problem unrelated to cibuildwheel, but numpy? See (and/or comment on) scipy/oldest-supported-numpy#45.

[neutrinoceros]

All I can say for sure is that the newest version of CBW fails the build, while 2.3.0 didn't. Maybe the internal setuptools upgrade is actually the problem.

[neutrinoceros]

reopening because the problem was addressed upstream with a new version of numpy shipping with Windows 32bits wheels, and a new version of oldest-supported-numpy to reflect it. Also the build time of this job is now 25% shorter.
I'll merge after all tests pass.

[neutrinoceros]

Actually just figured out how to trigger this workflow for PRs that change it, so we don't need to runs jobs manually or on forks to validate such an update. I won't self merge after all since this new patch wasn't reviewed yet.

[neutrinoceros]

switching to draft, will try to enable macOS arm builds while I'm at it

[neutrinoceros]

@matthewturk do you want to have another look at this ?