Evaluating the Fairness of Discriminative Foundation Models in Computer Vision