FIX Change MRO for nearest neighbors estimators

[alfaro96]

Reference Issues/PRs

Merge required to continue working on #17806.
Continuation from #14884.

What does this implement/fix? Explain your changes.

This PR changes the MRO for KNeighborsClassifier, RadiusNeighborsClassifier, KNeighborsRegressor and RadiusNeighborsRegressor.

Any other comments?

I have discovered a few more estimator to change MRO @rth.

[alfaro96]

I think we should move these lines:

scikit-learn/sklearn/neighbors/_base.py

Lines 401 to 402 in c3f2516

	else:
	X = self._validate_data(X, accept_sparse='csr')

before:

scikit-learn/sklearn/neighbors/_base.py

Line 1174 in c3f2516

return self._fit(X)

To:

• Avoid duplicate validation of data from nearest neighbors estimators that inherits from SupervisedIntegerMixin and SupervisedFloatMixin.
• Ensure that no error is raised in these lines:

scikit-learn/sklearn/base.py

Lines 410 to 415 in c3f2516

	if y is None:
	if self._get_tags()['requires_y']:
	raise ValueError(
	f"This {self.__class__.__name__} estimator "
	f"requires y to be passed, but the target y is None."
	)

when calling to _validate_data with nearest neighbors estimators that requires_y tag is set to True. Previously, no error was raised because requires_y tag was set to False due to wrong MRO.

[rth]

@alfaro96 The current change look good, but we would indeed need to fix tests. The proposed changes sound good to me (or at least it that be might be clearer once they are added to this PR) nevermind, I saw #17849

[alfaro96]

@alfaro96 The current change look good, but we would indeed need to fix tests. The proposed changes sound good to me (or at least it that be might be clearer once they are added to this PR) nevermind, I saw #17849

My fault, I forgot to comment that I was addressing this issue in #17849 :)

[alfaro96]

Closing since the MRO has been changed in #17849.