[LuceneOnFaiss - Part3] Added FaissHNSW graph.

[0ctopus13prime]

Description

In this PR, I've added FaissHNSW graph and a bridge that complies to Lucene's HNSW graph interface.
With the bridge, now we can use Lucene's HNSW graph searcher to perform vector search on FAISS index.

Related Issues

RFC : #2401

Check List

• New functionality includes testing.
• New functionality has been documented.
• API changes companion pull request created.
• Commits are signed per the DCO using --signoff.
• Public documentation issue/PR created.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[shatejas]

Will look at the unit tests soon.

Looks good otherwise, minor comments

[0ctopus13prime]

@jmazanec15
Hi Jack, please let me share the required memory space upper bound for long[] offsets within this naive approach, and the result how much we can compress the space with monotonic encoding.

Conclusion : With the encoding, required memory would be shrink to 475KB from 7.45GB for 1B vectors.

Current Approach
The number of vectors : 1B
Size of offsets required : 7.45GB

With DirectMonotonicWriter
Meta bytes : 163KB
Data bytes : 312KB
Total : 475KB = (163KB + 312KB)

Will make sure to add this encoding in the next PR.
Thank you!

[shatejas]

Looks good, need a clarification and its good to go

[shatejas]

just want to make sure that the vectors are not repeated on each level so we don't count extra

[0ctopus13prime]

there will be no duplicate counts as the index of levels (e.g. i) represents an internal vector id, and we increase the index sequentially.