Estimating Heterogeneous Treatment Effects Within Latent Class Multilevel Models: A Bayesian Approach