InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy