- Paginate heavy Cassandra reads (triples, graph/document embeddings)
using synchronous session.execute() in run_in_executor with fetch_size
paging, preventing materialization hang on large result sets
- Fix document stream endpoint to use workspace-scoped librarian queues
- Add decoder error handling for PDF/OCR/unstructured processors
- Add WebSocket mux guards for missing auth fields
- Add null check in librarian document streaming
- Rewrite get_document_content CLI to stream via librarian
- Add Poppler dependency to unstructured container