Add support for validation sets and callbacks on lgbm learner

[vitorsrg]

Status

READY

Todo list

• Documentation
• Tests added and passed

Background context

The lgbm_regression_learner did not support validation sets or callbacks, limiting the ability to use early stopping and monitor training progress — features already available in the classification counterpart (lgbm_classification_learner).

Continuation of #249 given dependencies unblocked resolution by #250

Description of the changes proposed in the pull request

• Add valid_dfs and callbacks parameters to lgbm_regression_learner, matching the interface of lgbm_classification_learner
• When valid_dfs is provided, create LightGBM validation datasets (with proper feature names and weights) for early-stopping support
• Pass callbacks through to lgbm.train for iteration-level hooks (e.g. lgbm.early_stopping, lgbm.log_evaluation)
• Update docstring to document the new parameters and fix an incorrect description of the target column
• Bumped coverage so it's compatible with newer versions of numba

Where should the reviewer start?

src/fklearn/training/regression.py — the changes to lgbm_regression_learner