Unveiling the Pitfalls of Knowledge Editing for Large Language Models