BPO: Towards Balanced Preference Optimization between Knowledge Breadth and Depth in Alignment