Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models