PLAT-3114: Keep only matched works when paging Crossref/DataCite
This MR stops run-kpi-report-script running the workers pod out of memory: fetch_work_uri_data_from_crossref() and fetch_work_uri_data_from_datacite() now pass each page through match_request_with_dois() as it arrives and keep only the publisher and missing books, instead of every full work record for the prefix. It also asks the APIs for just those fields (Crossref select=DOI,type,publisher,prefix, DataCite fields[dois]=doi,types,publisher), cutting each DataCite page from 4.9 MB to 0.3 MB.
Against the live APIs, old and new output are identical and peak memory per prefix drops from 225 to 8 MiB for 10.3726 (24,626 books) and from 104 to 4 MiB for 10.25932 (DataCite-only). The 16 Nov prod run was OOMKilled at the 1000Mi limit in process_data(), with five such prefixes in flight.