Towards 3D Acceleration for low-power Mixture-of-Experts and Multi-Head Attention Spiking Transformers