RTX PRO 6000 vs H100 for LLM Inference: Which Is More Cost-Effective in 2026?