Adding 3-ary PriorityQueue to speed up TopFieldCollector

### Description

In #15140, we introduced TernaryLongHeap to optimize performance for collectors such as TopDoc*Collectors. Nevertheless, TopFieldLeafCollector still adopts a 2-ary PriorityQueue, which causes identical performance bottlenecks even with a topN value of 1000.

Would it be feasible to implement a ternary-heap based PriorityQueue as well?

Start by comparing the existing TernaryLongHeap from #15140 with TopFieldLeafCollector and its current 2-ary PriorityQueue usage. Investigate the reported topN=1000 bottleneck and benchmark the ternary implementation against the existing queue; done means a validated PriorityQueue approach with improved collector performance.

Written by the indexing model from the issue text.