Extend the existing architecture to support JSONL dumps. This will allow for much more flexible creation of indices.