On Controller Tuning with Time-Varying Bayesian Optimization