SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Model