On the Calibration of Probabilistic Classifier Sets