Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models