Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs — ThinkLLM