Update comparison table: add BF16 variant, set BF16 as baseline

Upload BF16 TensorRT-LLM checkpoint for Qwen2.5-1.5B-Instruct

initial commit