Implement kmeans with kmeans++

[vberezhnev]

No description provided.

[ichorid]

Please add tests and remove CI/CD stuff (that is subject to a separate PR). Also, move the k-means algorithm into a separate file.

[vberezhnev]

@automainint What do you think about moving files like clustering.rs, log.rs, and astar.rs into a separate folder specifically for libraries?

[automainint]

@automainint What do you think about moving files like clustering.rs, log.rs, and astar.rs into a separate folder specifically for libraries?

I usually avoid adding abstractions without a specific reason. E.g. protocol.rs is a separate file because it's used by the psql-connector, while some other parts of the service are not. protocol.rs depends on log.rs, but not on astar.rs, so log.rs and astar.rs have distinct status, and it's not obvious that they have to be grouped together.

[automainint]

Each function that returns a score should also return the cluster value for that score. So we have to change function signatures, and it's probably better to apply (or read cached) clusters in functions fetch_score and fetch_all_scores.

Currently read_mutual_scores doesn't apply clustering correctly. All scores from a user node should be clustered, not all scores to a user node.

I will change function signatures and write tests in the new branch feature/score_clustering.

Also we will have to implement caching of clusters as "group bounds", as mentioned in #10 (comment) by @ichorid.