Evaluating Brain-Inspired Modular Training in Automated Circuit Discovery for Mechanistic Interpretability