ci: Test packaging flow on Windows

[ichard26]

Most of pip's maintainers use non-Windows environments, so it's easy for our tooling to regress on Windows. Let's avoid that for our release flow.

In addition, remove the dependency on the packaging job by the test jobs. The packaging job rarely fails so we aren't saving any CI resources by failing early if the packaging job fails.

The only reason to gate the tests jobs on the packaging step would be to reduce CI failure noise when pip's packaging is broken, resulting most of the tests jobs also failing. By gating the test jobs on the packaging job, only the packaging failure would be visible, making easier to figure out the root problem. I don't think this happens much in practice though.

[ichard26]

On second thought, I'm not sure how much value this adds given this won't catch subtle failures,¹ but we do have a custom build script now so perhaps that's enough surface area to warrant additional testing. cc @pfmoore

Footnotes

1. Unless we compare the artifacts across platforms, which would be better done when we fully support deterministic builds of pip (which is blocked on setuptools). ↩

[ichard26]

If no one objects in the next few days, I'll merge this. This probably has limited benefit, but if it avoids release-day chaos because of Windows (even merely due to a silly and trivially easy to fix mistake), it's worth it IMO.