Score given and middle separately in the agreement records, where a source states the split #702

cldr_agreement.py scores the given name and the middle names as one given_names role. So a parse that splits a name wrongly passes. Kim Min Ho under FAMILY_FIRST reads given Min, middle Ho, though Min Ho is one given name, and the conventions corpus records it as agreeing. The same holds for Lim Mei Ling Jessica and Tran Thi Mai.

Proposal: score given and given2 separately, only for sources that state the split: CLDR (given2, 1,305 rows) and the conventions corpus. GND and Wikidata record every given name in one field, so they keep the merged role.

Measured on master 9e7b7280: among rows whose merged role already agrees, a split view would newly disagree on 265 CLDR rows and 3 conventions rows. GND (128), Wikidata (7) and Wikidata native (8) would disagree too if scored, which is why they must not be. Most of CLDR's 265 come from the parser's convention (first word given, the rest middle) against CLDR's multi-word given (Cornelia Margarita), not from a bug.

So: add the split as a report-only view first, like the LC cross-check in #693, rather than a ratcheted role. Ratchet it once the design settles what middle means for compound given names (María José, Min Ho, Muhammad Salim).

Open questions:

• What does middle mean when a culture's given name is several words? (This is design work, not scorer work.)
• Should the conventions corpus restate rows like Tran Thi Mai (Thi as given2) once that's decided?