Simplifying Adversarially Robust PAC Learning with Tolerance