✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: a5f3149. Link to the linter CI: here}

What is the reason why this is an issue for the partial dependence but not in other places?

It is also a problem in other places, either when randomizing the order of the tests (as discovered in #29584) or when using parallel threads, as investigated in #30007.

To fully solve #30007 we would need to fix all occurrences of such side effects caused by mutable pytest parametrizations but also, avoid running some tests that are fundamentally thread-unsafe (e.g. tests that check warnings).

This will be an incremental maintenance effort, and I decided to open a small PR to resolve a single problem at a time in a single test module.

I don't feel super good about this ... this is a typical pytest pattern and I feel like in an ideal world we should not have to change our tests too much to stress-test free-threaded (probably easier said than done ...). I guess adding .clone is not a big change but we are going to add it in many many places just for free-threaded reasons.

Note that this is slightly different than #29584 which happens because the same list of mutable objects is reused to parametrize different tests. That means side-effects in one test can impact the next test that run and that the ordering of tests may matter. This issue is specific to pytest-run-parallel (or similar plugin like pytest-freethreading) which runs the same test function in multiple threads with the goal to find issues when using free-threaded.

we are going to add it in many many places just for free-threaded reasons

Maintaining a side effect free test suite is helpful to ease debugging in general, even if I understand that there is a difference with what is being fixed in #29584.

The alternative would be to require the .fit method of scikit-learn estimators to be thread-safe, but this would be a much more complex endeavor.

I think this is a good fix regardless of @lesteve 's point, and those fixes can happen later?