ENH Add Array Api compatibility for mean_squared_error

[Tialo]

Towards #26024

Reference Issues/PRs

What does this implement/fix? Explain your changes.

Any other comments?

[github-actions]

✔️ Linting Passed

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

_{Generated for commit: a2d5a34. Link to the linter CI: here}

[OmarManzoor]

LGTM. Thanks @Tialo.

I verified that the tests pass with cuda and mps devices.

[ogrisel]

LGTM but let's see if the tests pass on CUDA GPUs with the following manually triggered CI run on our new CUDA CI infrastructure:

https://github.com/scikit-learn/scikit-learn/actions/runs/9399022364

[ogrisel]

For information, I tested locally with on Apple M1 to check the device="mps" case and the test passes:

sklearn/metrics/tests/test_common.py::test_array_api_compliance[mean_squared_error-check_array_api_regression_metric-torch-mps-float32] PASSED

[ogrisel]

I indeed forgot to merge main before triggering the CUDA GPU CI. Let me try again with the new state of the PR:

• 9400531472

[ogrisel]

LGTM CUDA tests are green, thanks for the PR.