Machine Learning Time Propagators for Time-Dependent Density Functional Theory Simulations