From Spatial to Actions: Grounding Vision-Language-Action Model in Spatial Foundation Priors