Unified Discrete Diffusion for Simultaneous Vision-Language Generation