One embedding unit produced from a ContentItem: a slice of its text plus the id minted for the
chunk up front. chunkID becomes BOTH the ContentItemChunk row's PK and its vector-DB id (under
the 'recordId' strategy), so it is known at metadata-build time — which lets a chunk vector carry
its own identity (buildVectorMetadata) even though the chunk row is written later.
One embedding unit produced from a ContentItem: a slice of its text plus the id minted for the chunk up front.
chunkIDbecomes BOTH the ContentItemChunk row's PK and its vector-DB id (under the 'recordId' strategy), so it is known at metadata-build time — which lets a chunk vector carry its own identity (buildVectorMetadata) even though the chunk row is written later.