A Framework for Testing Identifiability of Bayesian Models of Perception