Diffsound: Discrete Diffusion Model for Text-to-Sound Generation