Calibrating Bayesian Learning via Regularization, Confidence Minimization, and Selective Inference