Thinking Inside the Mask: In-Place Prompting in Diffusion LLMs