Rethinking Out-of-Distribution Detection From a Human-Centric Perspective