Don't Take the Premise for Granted: Evaluating the Premise Critique Ability of Large Language Models