Make LVLMs Focus: Context-Aware Attention Modulation for Better Multimodal In-Context Learning