[CI] PyPy tests failing lately and it is difficult to identify the root cause

[abravalheri]

I first noticed this behaviour when reverting the deprecations introduced in v78.0.0.
But lately we have seen PyPy failing in what seems to be unrelated issues.

Example extracted from #4931:

We can see an error in test_setupcfg.py:488 which does not really have a good explanation.
In the setuptools.tests.config.test_setupcfg:TestOptions.test_basic function, we use the fake_env, but by itself it does not add any classifier. So we have a warning about classifiers without having specified any classifier?

[abravalheri]

The fact that the error keeps changing seems to me that we have some sort of race condition,
specially when it comes to the tmp_path/tmpdir fixture. (It seems that the temp directory is "reminiscing" from one test to the other)

[abravalheri]

In #5144 we can see the PyPy tests failing again where it is difficult to imagine why.

Also I noticed that the tests that is actually failing keeps changing from run to run:

In this CI run, the only error-ed function was test_warn_accidental_...

FAILED setuptools/tests/config/test_setupcfg.py::TestOptions::test_warn_accidental_env_marker_misconfig[[options.extras_require]\nfoo = bar;python_version<3\n] - Failed: DID NOT WARN. No warnings of type (<class 'setuptools.warnings.Setu...
= 1 failed, 1574 passed, 37 skipped, 9 xfailed, 3 xpassed in 674.52s (0:11:14) =

In the successive CI run (after an xfail is applied), another function fails:

FAILED setuptools/tests/config/test_setupcfg.py::TestOptions::test_raises_accidental_env_marker_misconfig[[options]\ninstall_requires = bar;python_version<'3'\n] - Failed: DID NOT RAISE <class 'packaging.requirements.InvalidRequirement'>
= 1 failed, 1570 passed, 37 skipped, 9 xfailed, 7 xpassed in 590.81s (0:09:50) =
py: exit 1 (595.06 seconds) /home/runner/work/setuptools/setuptools> pytest pid=18734

When the PR was merged, yet another error appears:

And then, just by clicking the "re-run jobs" button, the CI mysteriously passes...

(this seems to corroborate the theory that there is some form of race condition affecting how the tests are running)...