[FEATURE] Support filter queries for approximate k-NN that is based on Faiss

[martin-gaievski]

Is your feature request related to a problem? Please describe.
We do not have support to effectively filter docs and then apply k-NN specifically on the filtered docs. This will help guarantee to obtain k Nearest Neighbors on the filtered set. Please note, we're requesting only for Faiss based approximate k-NN search.

Describe the solution you'd like
Ideally we'd like to do k-NN approximate search on a sub-set of docs that match filtering criteria and have a guarantee of top K docs in result. We want to be able to define the filter(s) in knn query and use existing OpenSearch queries as filters (boolean, term etc.).

Describe alternatives you've considered
Pre-filtering is supported with scoring script but in form of exact brute-force k-NN search. Post-filtering can be done as approximate search with post-processing, but problem is that top k results are not guaranteed.

Additional context
We'd like to have parity with k-NN powered by Lucene, opensearch-project/k-NN#376

[dblock]

Should this issue live in the k-nn repo?

[martin-gaievski]

Should this issue live in the k-nn repo?

agree, let me close this one