MB-59102: Merging kNN hits from a distributed index with exact search hits

[CascadingRadium]

• Reverts commit MB-59102: Merging KNN Results #1910, which was an earlier attempt to address this issue.
• Implements the PreSearch Construct in Bleve alias search, enabling a preliminary query to collect metadata from all alias indexes before executing the main search query in MultiSearch. PreSearch gathers KNN results from all alias indexes, selecting the top K results. This facilitates the main Bleve Query to operate within the context of documents that matched the KNN query, ensuring seamless functionality of existing Bleve constructs such as Faceting, Sorting, Pagination, SearchAfter, and SearchBefore.
• Introduces the KNN Collector construct to merge and obtain accurate Top K results from multiple Zap Segments' KNN results.
• Enhances KNN Unit Tests for greater generality.
• Addresses an issue where errors generated within the Top N Document handler were being discarded.
• Resolves an issue where document matches failing to meet the SearchAfter clause weren't being returned to the pool.

[abhinavdangeti]

Thank you @CascadingRadium . I'll share my review by tomorrow.
Meanwhile - @Thejas-bhat , @metonymic-smokey , @moshaad7 , @Likith101 would you please start reviewing this.

[abhinavdangeti]

@CascadingRadium I tried this change E2E with couchbase. Things don't work on 2 accounts -

• one, if I don't specify knn_operator, I see unknown knn operator: - this is not right, it needs to default to or when unspecified
• two, I go ahead and put in knn_operator to "or", a kNN search over that SIFT dataset in combination with match_none, now returns nothing (it seems the response is just empty)

Non-vector indexes are affected as well, so this needs a bit more testing from your end.
Can you look into what's going on? Should be straightforward to reproduce.

[metonymic-smokey]

@abhinavdangeti Re point 1, I believe this PR is the cause(#1935) since this doesn't allow an empty string as the KNN operator. If it is, it should be reverted.

[abhinavdangeti]

Thanks @metonymic-smokey , here's the PR adjusting that - #1937

[CascadingRadium]

there's a separate PR in cbft that I will raise today. the marshal unmatshal in logic there needs to be updated

[abhinavdangeti]

there's a separate PR in cbft that I will raise today. the marshal unmatshal in logic there needs to be updated

I don't see this still. Are any gRPC changes required in cbft to support this?

[Thejas-bhat]

still trying to understand the details of the implementation, although i get the core logic behind it. will keep at it periodically and let you know any concerns i have

[abhinavdangeti]

Let's go with this. Well done @CascadingRadium .
Keep an eye out for any optimizations we can fish out in this area for the future.