Instead of one result file line per document, add them in batches
to reduce the overall number of documents in the collections.
Each document now contains a `result_lines` entry, which is a list
of dictionaries, one result file line per list element.