Evaluating the Fairness of Deep Learning Uncertainty Estimates in Medical Image Analysis