The influence of trial order on learning from reward vs. punishment in a probabilistic categorization task: experimental and computational analyses