[3b] Async ingestion worker — chunk, embed, index

[andrmaz]

What to build

Implement the BullMQ worker that consumes ingestion jobs enqueued by source upload. The worker normalizes raw content, splits it into text chunks, generates embeddings via the configured embedding provider, and persists Document + Chunk rows with pgvector embeddings and org ownership metadata. Failed jobs transition to an error state with a surfaced reason.

Acceptance criteria

• Worker picks up enqueued jobs and processes them without blocking the request path.
• Chunks and embeddings are persisted in the Chunk table with pgvector column populated.
• Document and Chunk rows carry Organization FK.
• A failed job (e.g. embedding provider error) surfaces an actionable error state on the source record.
• Integration test covers the happy path: upload → job → indexed chunks.

Blocked by

• [3a] Source registration + file upload #6 [3a] Source registration + file upload