add a parameter to support "interactive constraints"

[pedwards-baasco]

The upstream dmlc xgboost has a feature called interaction constraints.

This feature is useful to train highly explainable models for high-risk applications like lending. It would be wonderful if TFDF boosting supported a similar option.

[janpfeifer]

Per offline chat, for others interested, this is a simple and almost equivalent approach, by ensembling multiple individual trees, each trained with a subset of features (only one feature in the example):

https://colab.research.google.com/drive/17LEDCwsf1-x2cBKz0J43SES8EDeiy-AB?usp=sharing

Maybe this will be helpful for others, while we don't implement the feature.