Quamba: A Post-Training Quantization Recipe for Selective State Space Models