Muse: Text-To-Image Generation via Masked Generative Transformers