Try using Lucene IntIntHashMap to speedup and reduce memory usage of top-K counting

[alexklibisz]

Background

Breaking this out of #160, specifically this comment: #160 (comment)

I'd like to try using the IntIntHashMap from Lucene 9.x to decrease latency and memory usage for counting the top-K hits in LSH queries.

Deliverables

• Implement top-k counting using IntIntHashMap
• Demonstrate that it's a measurable speedup in benchmarks

Related Issues

So far it looks like the speedup is only noticeable in larger datasets. So this might require expanding benchmarking to datasets larger than Fashion Mnist.

[alexklibisz]

There are also some other interesting classes in Lucene that seem related, e.g., RoaringDocIdSets.