Speedup internal CI

[EwoutH]

I've noticed cibuildwheel internal CI takes quite long, which makes iterating quickly more difficult and time-consuming. I was curious if we have some ideas to speed some of those jobs up. Currently the jobs that take the longest:

• The test matrix job of the GitHub Actions test.yml workflow (~40 to ~45 minutes)
• The windows_38 job of the Azure DevOps azure-pipelines.yml (~70 to 90 minutes)

There might be some parallelization or other optimizations possible here to reduce the runtime of these tasks.

For a project like cibuildwheel, I also wouldn't find it the strangest to contact Microsoft to see if there's something possible in more powerful hosted CI runner hardware.

For example, in the macos_38 job of the Azure pipeline, parallelization is already used with --num-processes 2. Can we get runners with more cores on Linux and Windows as well?

Or could we split up the test set into smaller chucks each running on a separate runner?