Mitigating the Language Mismatch and Repetition Issues in LLM-based Machine Translation via Model Editing