[Feat] KNN filtering with limit and KNN distance function

[emmanuel-keller]

What is the motivation?

KNN operations should be able to be filtered, as described in #4000.

What does this change do?

This PR modifies the behavior of the KNN search so that the search reaches the K goal by bypassing documents that do not match the query.

It also introduces the vector::distance::knn() function, which returns the distance computed during the query (avoiding recomputation).

Eg.:

SELECT id, vector::distance::knn() AS distance FROM pts
WHERE flag = true && point <|2|> [4.5f] ORDER BY distance;

• HNSW implementation
• MTree implementation
• Bruteforce implementation

What is your testing strategy?

Tests have been written (for MTree and HNSW).

Is this related to any issues?

Fixes #4000

Does this change need documentation?

If this pull request requires changes, updates, or improvements to the documentation, then add a corresponding issue on the docs.surrealdb.com repository, and link to it here.

• Documents vector::distance::knn docs.surrealdb.com#558

Have you read the Contributing Guidelines?

• I have read the Contributing Guidelines

[phughk]

Nice, really cool thanks!

[phughk]

Lgtm thanks!