Assessing the Interpretability of Programmatic Policies with Large Language Models