[NeurIPS‘25 Spotlight] Code for the paper "Mozart: Modularized and Efficient MoE Training on 3.5D Wafer-Scale Chiplet Architectures"