Regularization Guarantees Generalization in Bayesian Reinforcement Learning through Algorithmic Stability