Unifying Human Motion Synthesis and Style Transfer with Denoising Diffusion Probabilistic Models