Rubrics as Rewards (RaR): A Reinforcement Learning Framework for Training Language Models with Structured, Multi-Criteria Evaluation Signals