CATS: Contextually-Aware Thresholding for Sparsity in Large Language Models