Establishing trust in decisions made from data: Physics-informed machine-learning models with computable generalization bounds