HAMP-LIC: Hessian-Aware Mixed-Precision Post-Training Quantization for Learned Image Compression