Bayesian optimization under mixed constraints with a slack-variable augmented Lagrangian