Continuous-in-time Limit for Bayesian Bandits