Overhaul socket stubs on Windows and MacOS

[AlexWaygood]

Helps with #8101.

This PR reduces the (_)socket allowlists for win32 and darwin to only a handful of missing constants/functions.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[hauntsaninja]

Thanks, this is a pretty heroic PR. I'd want to double check there aren't any Linux regressions before merging.

Can confirm stubtest also looks good on my Mac laptop (in case there were OS version or hardware specific things we're missing). For what it's worth, I have socket.MSG_NOSIGNAL on 3.8 and above — curious what you were seeing.

[AlexWaygood]

For what it's worth, I have socket.MSG_NOSIGNAL on 3.8 and above — curious what you were seeing.

I was using GH Actions on my fork to run stubtest on macos, and it seemed like it varied from one stubtest run to the next as to whether it existed at runtime.

[AlexWaygood]

I'd want to double check there aren't any Linux regressions before merging.

Here's a failing Ubuntu stubtest run from yesterday on my fork, before I added the allowlist back for linux, if that's helpful: https://github.com/AlexWaygood/typeshed/runs/6955971392?check_suite_focus=true

I'd rather not make any linux-specific changes in this PR, if at all possible, since (1) I don't have a linux machine, and (2) I know very little about linux and the various linux operating systems.

[hauntsaninja]

Confirmed that we didn't accidentally regress availability of any constants on Linux on Python 3.9 (same set of "is not present in stub" errors)

[AlexWaygood]

Confirmed that we didn't accidentally regress availability of any constants on Linux on Python 3.9 (same set of "is not present in stub" errors)

Awesome, thanks for checking!