Grounding by Trying: LLMs with Reinforcement Learning-Enhanced Retrieval