Spatial Forcing: Implicit Spatial Representation Alignment for Vision-language-action Model