Add feature_mask to GLM

[billbrod]

Right now, only PopulationGLM has a feature_mask attribute. But feature_mask could also be used by GLM to select which inputs are necessary (as done in https://github.com/flatironinstitute/ccn-software-jan-2025/, the model selection notebook).

This would involve moving all feature_mask logic (including __sklearn_clone__, introduced in #291 ) to GLM. Or does it belong in BaseRegressor?