Personalized LLM Decoding via Contrasting Personal Preference