[FEATURE] Implement parallel execution of sub-queries for hybrid search

[martin-gaievski]

Is your feature request related to a problem?

Currently individual queries of Hybrid query (aka sub-queries) are executed sequentially (Initial implementation that is done under https://github.com/opensearch-project/neural-search/issues/123). That may be sub-optimal as system may waste time on waiting.

Obvious approach for such execution is to run every sub-query execution using parallel thread and wait for results from all thread, this will be at shard level. Results then will be combined and sent to coordinator node all at once, as it's done today.

What solution would you like?

Sub-queries may be executed in parallel. Actual improvement must be verified by running benchmarks on sequential and parallel approaches.