Mislearning from censored data: The gambler's fallacy and other correlational mistakes in optimal‐stopping problems