TST: Seemingly non-deterministic order of value_counts #32514
Description
We had a flaky test for test_value_counts in pandas/tests/base/test_ops.py which was related to the sort order of entries with the same count. A hotfix was introduced in #32449 and resolves the CI issue.
The underlying issue still isn't fixed/confirmed. The test assumes that in case of same counts the order of the original index/series object is preserved. This assumption breaks on CI though. Please refer to my comment (and the following ones) for all related details I have found so far.
It doesn't necessarily seem to be related to the narrow dtypes. It's probably just more likely that values are duplicated the narrower the dtype is.