ENH Add array_api compatibility to max_error

[EdAbati]

Reference Issues/PRs

Towards #26024

What does this implement/fix? Explain your changes.

It makes the max_error implementation compatible and tested with the Array API.

[github-actions]

✔️ Linting Passed

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

_{Generated for commit: 0e88c7d. Link to the linter CI: here}

[EdAbati]

Had to change the type of y_true in the regression test from int to float because of this

Was there a reason why that was int?

[EdAbati]

mps and CUDA tests are green on my side :)

[ogrisel]

LGTM.

[ogrisel]

Thanks again for you Array API PRs @EdAbati.

Since you already had quite a bit of experience on those I was wondering if you wouldn't be interested in contributing on PRs that are a bit less well defined but potentially more impactful from a performance point of view?

For instance I think we could improve the array API support for the solver="cholesky" case of the Ridge class by leveraging the xp.linalg.solve standard API.

[EdAbati]

Hi @ogrisel , sounds good!

I was also looking at dcg_score that, after a brief look, seems to require things are not implemented yet.
I will also make a draft PR with that, so we can discuss

[betatim]

I triggered the CUDA CI for this PR https://github.com/scikit-learn/scikit-learn/actions/runs/9500016241

(Mostly for my own sake to use the CI a bit)