DiffiT: Diffusion Vision Transformers for Image Generation