Optimal Dynamic Regret in LQR Control