Investigating Multi-Hop Factual Shortcuts in Knowledge Editing of Large Language Models