Stimulus-Dependent Adjustment of Reward Prediction Error in the Midbrain