make training features configurable

[jpata]

For all 3 models, both input_dim of each model and the unpack_... functions (https://github.com/HEP-KBFI/ml-tau-en-reg/blob/main/enreg/scripts/trainModel.py#L46-L71) need to take into account three possible feature sets:

• only kinematics
• kinematics + additional features like PDGs
• kinematics + additional features + lifetimes

This should be easily configurable with a single flag (e.g. feature_set=[kinematics], feature_set=[kinematics,pdgs], feature_set=[kinematics,pdgs,lifetimes], to enable scanning across the options in https://github.com/HEP-KBFI/ml-tau-en-reg/issues/17.

[jpata]

possible next task for @HardiVv.