DOC: Note alpha_per_target incompatibility with cv in RidgeCV

[shouhamdi]

Summary

The alpha_per_target parameter of RidgeCV raises a ValueError when
used together with cv != None, but the docstring does not mention this
restriction. This PR adds the same one-line note already present on the
sibling parameter store_cv_results:

This flag is only compatible with cv=None (i.e. using
Leave-One-Out Cross-Validation).

Evidence

• Runtime guard in _BaseRidgeCV.fit():

  if self.alpha_per_target:
      raise ValueError("cv!=None and alpha_per_target=True are incompatible")

• Existing test in test_ridge.py (test_ridge_gcv_vs_ridge_loo_cv) confirms
  this ValueError is expected for both cv=LeaveOneOut() and cv=6.
• The parallel parameter store_cv_results already documents an identical
  constraint — this PR brings alpha_per_target in line.

Change

Two lines added to the alpha_per_target docstring in RidgeCV
(sklearn/linear_model/_ridge.py). No code or test changes.

Made with @tomMoral and @kingjr

[AnneBeyer]

Thank you @shouhamdi for spotting this. LGTM!

[Abdeltoto]

Thanks, this looks correct to me. I checked that the added note matches the existing \ValueError\ path for \�lpha_per_target=True\ with \cv is not None, and it is consistent with the neighboring \store_cv_results\ documentation.

[GaelVaroquaux]

LGTM.

Thanks a lot @shouhamdi for the PR, and @AnneBeyer and @Abdeltoto for the reviews.

Merging, as such a DOC PR needs only one core-dev review