Understanding the scalability of Bayesian network inference using clique tree growth curves