Bayesian Domain Invariant Learning via Posterior Generalization of Parameter Distributions