Inference-Time Scaling

C$^2$: Scalable Auto-Feedback for LLM-based Chart Generation