On the Role of Calibration in Benchmarking Algorithmic Fairness for Skin Cancer Detection