Compatibility with scikit-learn 1.6rc1

[FBruzzesi]

Description

Diff is simpler than expected:

• Fix order of mixin's inheritance
• Added the new __sklearn_tags__ to all the classes that changes the default tags
• Created a utility function in sklego.common that perform validation either via validate_data (scikit>=1.6) or check_X_y and check_array (scikit < 1.6).

There are some checks which are not fully compatible, and one that might be buggy in this RC: check_sample_weight_equivalence seems to fail on every of our estimators that support sample_weight - it might be worth reporting upstream, but I could not pin the issue

Closes #719

[FBruzzesi]

So current tests are all passing, which is definitly a good sign that we did not break anything. I tested with scikit-learn1.6rc1 locally (python3.10, ubuntu) and successfully

[koaning]

Whoa. So many changes.

I guess the best check is to merge it now and to see if something breaks in the future, but nothing jumped out of the ordinary.

[koaning]

Ah, I did the merge and now stuff seems breaking again 😓

[koaning]

Changing state to reflect some tests break.

[koaning]

This is ... odd. The tests locally in codespaces run just fine but they break in CI.

[FBruzzesi]

I will try to take a better look this weekend.. Should we pin scikit-learn version until we fix this?

[koaning]

Ah, 1.6 is no longer a release candidate I think.

Let's delay version pinning unless folks start to submit issues. It could be that we are worrying about details that aren't a huge worry for folks.

[FBruzzesi]

Yes, 1.6 is now definitly released 🙃