Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization