Describe the N+1 query problem in aggregation. How to efficiently count related objects, such as the number of comments on 100 posts, without loading all related entities into Python memory?