A Consequentialist Critique of Binary Classification Evaluation Practices