When Claims Evolve: Evaluating and Enhancing the Robustness of Embedding Models Against Misinformation Edits