[python-package] change types in sklearn.py initializations

[jameslamb]

Contributes to #3756.
Contributes to #3867.

Follow-up to #5672 (thanks @IdoKendo for starting this discussion again!)

Fixes the following errors from mypy.

python-package/lightgbm/sklearn.py:897: error: Incompatible return value type (got "None", expected "int")  [return-value]
python-package/lightgbm/sklearn.py:904: error: Incompatible return value type (got "None", expected "int")  [return-value]
python-package/lightgbm/sklearn.py:918: error: Incompatible return value type (got "Optional[int]", expected "int")  [return-value]

See my description in #5672 (comment) for the reasoning behind this proposed change.

How I tested this

Besides just running mypy and the unit tests, I wanted to also confirm that there wasn't any other code relying on the attributes _n_features, _n_features_in, and _best_iteration being initialized to None.

Since those are private attributes of a class inside lightgbm, it's sufficient just to check this repository for such references.

git grep '_n_features'
git grep '_best_iteration'

I didn't see any places relying specifically on those attributes being initialized to None, so I think changing them to be initialized to -1 is safe.

[jameslamb]

I'll wait to merge this one until you have a chance to also review, @jmoralez (no rush!)

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed.
To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues
including a reference to this.