WIP: CI: Pameterize locale tests specifically for Linux

[mroeschke]

Looking to eventually eliminate the local-specific builds by parametrizing in specific tests

[burnpanck]

What are the motivations which have led to considering to remove the locale-specific builds? The reason I ask is because I see two different reasons to test against multiple locales:

1. Test intended locale-specific behaviour
2. Prevent unintended locale-specific behaviour

The first category will probably benefit from this PR, as testing multiple locales becomes much faster and thus can be performed more frequently and efficiently. However, the second category will effectively lose coverage.

Specifically, I personally run on the locale de_CH, where multiple tests cause false positives. That makes impractical for me to run the test suite locally, so all my contributions are blind. I was about to open a PR to include de_CH in the test-set, as a basis to provide contributions to fix those tests. Should I do that anyway, or how would you like me to proceed with that issue?

On another note, just to double-check: The with_locale fixture and the tm.set_locale context manager it uses are neither thread-safe nor context safe. They set the locale process-wide. This could cause trouble in some parallel testing scenarios, although I believe pytest-xdist does spawn separate processes before running any test, so the sub-processes should have independent locales.

[burnpanck]

See also #44625 and #45013.

[mroeschke]

Currently pandas only tests a European locale it_IT and non-latin locale zh_CN on PY38 (in 2 separate builds)

This PR ensures we are testing these locales (also with en_US) on all our Linux PY38/39/3.10 builds so locale testing is being expanded not removed.

After this PR it would be easier in theory to tests more locales like de_CH. And pyetest-xdist runs tests in multiple processes so hopefully the thread safety concerns should not be an issue here.