Posterior Sampling-Based Bayesian Optimization with Tighter Bayesian Regret Bounds