Vector index table contains too many rows

[kungasc]

ALTER TABLE xxxxx
ADD INDEX idx_vector2
GLOBAL USING vector_kmeans_tree
ON (embedding)
WITH (distance=cosine, vector_type="float", vector_dimension=1024, levels=2, clusters=200);

SELECT COUNT(*) FROM `xxxxx`;

column0
288624165

SELECT COUNT(*) FROM `xxxxx/idx_vector2/indexImplPostingTable`

column0
290170401

[vitalif]

Two problems caused this bug:

1. KMeans sample collection was repeated after a schemeshard restart when it was already finished and uploaded into the level table.
2. KMeans samples were persisted incorrectly: new samples were added, but extra samples removed from buildInfo.Sample.Rows were not removed from the DB.