When Rubrics Fail: Error Enumeration as Reward in Reference-Free RL Post-Training for Virtual Try-On