Learning the Riccati solution operator for time-varying LQR via Deep Operator Networks