Support many-to-one in output transform modals for ML search response processors

[ohltyler]

ML search response processors uniquely allow to process one document at a time, or all at once (see one_to_one parameter in the docs here). We should customize the fetched data, and how the transforms work, depending on if one_to_one is true/false. Currently, it is showing as one-to-one, as that is the only experience for the other 2 ingest and search request contexts.