Rec-R1: Bridging Generative Large Language Models and User-Centric Recommendation Systems via Reinforcement Learning