First-Principles Study Combined with Interpretable Machine-Learning Models of Bayesian Optimization for the Design of Ultrawide Bandgap Double Perovskites