TSPORec: Token Selection via Preference Optimization for LLM-Based Sequential Recommendation