We've seen that some tests (at least the backend conformance test) was only executed by the test-win CI job.

Additionally to improving that, I'd like to regularize the cross-platform tests to make it more straightforward to add or change runners; and also with the goal of not having more "one-off" runners than necessary, as that makes it easier to see if the trigger for a failure is the specifics of the test configuration or something else (i.e., platform).

• Move win test runner to tox
• Unify win&osx runners in a CI test matrix. pyodide is a special snowflake, so will not be part.
  • Fix or skip resulting osx failures seen here
  • Choose a shorter name than "test-cross-platform", due to github length limits in display (see). "cross" (see)?
• Verify that all relevant test dirs are executed somewhere in the standard test matrix "test"
  • tests/crosshair conformance test (added to niche)
  • De-flake conformance test (see)
• Update the list of github "required" tests

See thread at #4536 (comment) for initial discussion