Learning Physics from the Machine: An Interpretable Boosted Decision Tree Analysis for the Majorana Demonstrator