CreditR1: Calibration-Aware Reinforcement Learning for Interpretable Corporate Credit Risk Assessment with Large Language Models | Synapse