Lifelong Personalized Low-Rank Adaptation of Large Language Models for Recommendation