Cache-DiT: A PyTorch-native Inference Engine with Hybrid Cache Acceleration and Massive Parallelism for DiTs.