On Pruning for Score-Based Bayesian Network Structure Learning