Fast interior-point inference in high-dimensional sparse, penalized state-space models