Optimizing Token Usage on Large Language Model Conversations Using the Design Structure Matrix