PEER: Unified Process-Outcome Reinforcement Learning for Structured Empathetic Reasoning