Scaling Diffusion Transformers Efficiently via $μ$P