SplitQuantV2: Enhancing Low-Bit Quantization of LLMs Without GPUs