Learnable Privacy Neurons Localization in Language Models