Decoding Large Language Diffusion Models with Foreseeing Movement