Quamba2: A Robust and Scalable Post-training Quantization Framework for Selective State Space Models