Text-Conditioned Sampling Framework for Text-to-Image Generation with Masked Generative Models