Assessing Bayesian Semi‐Parametric Log‐Linear Models: An Application to Disclosure Risk Estimation