How to support query cache in knn search

### Description

Currently, knn search does not use LRUQueryCache for acceleration. The knn search is heavy when k is big.
Can query caching be supported in knn search?

In semantic search scenario, there are many hot queries. Query caching can speed up these queries.

Start by tracing the knn search path and the existing LRUQueryCache integration. Determine whether repeated semantic-search queries can be cached safely, including the effect of large k values; done means supported hot queries use the cache without changing search results.

Written by the indexing model from the issue text.