Unifying Scene Representation and Hand-Eye Calibration with 3D Foundation Models