Delete all chunks for a source file (called before re-chunking a changed file).
Remove staging rows after successful upload — SQLite is a pure staging buffer.
Returns all rows that have a dense vector (excludes metadata-only rows).
Returns all chunk rows as Chunk objects, regardless of embedding or upload status.
Dense-text hashes of chunks for a file that already have a dense vector.
Returns a file → revision map from file_revisions only (excludes chunks table).
Returns a file → revision map for all files tracked in the DB.
Alias for getPendingUploadChunks() — keeps Uploader compatible.
Chunks inserted during chunking that still need a dense vector computed.
Embedded chunks that have not yet been uploaded to the vector store.
Insert fully-embedded chunks (used by batch pipeline callers).
Insert chunk metadata only (no dense vector yet). Safe to call inside a transaction.
Count of embedded chunks not yet uploaded.
Atomically replace all chunks for a source file with a new set, preserving existing dense vectors for chunks whose denseTextHash hasn't changed. fileRevision is the git blob SHA; required when chunks may be empty.
OptionalfileRevision: stringStore the dense vector for a previously inserted chunk.
Clear dense vector after upload to reclaim storage.