Near Universal Consistency of the Maximum Pseudolikelihood Estimator for Discrete Models