On the Impact of Random Seeds on the Fairness of Clinical Classifiers