[FEATURE] Asymmetric Shapley Values

It would be great if TreeShap function below can support Asymmetric Shapley Values (ASV)

RegTree::TreeShap (xgboost/tree_model.cc), using causal relationship as inputs, the weights of features in TreeShap are adjusted.

ASV paper:
[Asymmetric Shapley Values](https://www.researchgate.net/publication/336577536_Asymmetric_Shapley_values_incorporating_causal_knowledge_into_model-agnostic_explainability)

Start by reading RegTree::TreeShap in xgboost/tree_model.cc and the linked Asymmetric Shapley Values paper. Clarify how causal relationships should be supplied and how they should adjust TreeShap feature weights; done means the agreed ASV behavior is supported there.

Written by the indexing model from the issue text.