IROTE: Human-like Traits Elicitation of Large Language Model via In-Context Self-Reflective Optimization