Ignore Similarity If You Can: A Computational Exploration of Exemplar Similarity Effects on Rule Application