[ENHANCEMENT] query caching for vectors

[jn2clark]

Is your feature request related to a problem? Please describe. for some search patterns, the queries will be from a non-uniform distribution and a small number of queries will make up a disproportionate amount of all queries. caching the embeddings can improve throughput

Describe the solution you'd like some caching mechanism for search queries. the query could be vectorised and then cached. the type of cache is tbd but if it is large enough the priority could be based on counts or a combination of time and counts