MAINT define a single time _estimator_has and refactor code

[glemaitre]

From past discussion, I realized that we are defining the same _estimator_has in several places while it does exactly the same job and has the same semantic.

I think we should do a bit of cleaning by moving this function into a submodule in sklearn.utils. I would probably keep this function private for the moment even thought it could be useful for developers of third-party libraries.

Maybe @StefanieSenger would be interested in leading the effort since you should be familiar with this function after working on #28167?

[StefanieSenger]

Oh yes, that's a good effort and a nice project to work on. Thank you, @glemaitre, I will open a PR for this.