pub fn search_ids<C, R, S>(
query: &[(DimId, Weight)],
ids: &[RecordId],
cursors: R,
sink: S,
scratch: &mut Scratch,
) -> Vec<(RecordId, f32)>Expand description
Score only ids — sorted ascending, without duplicates — by seeking
every lane to each of them, and offer the ones present in at least one
lane to the sink. Cost O(|ids| × lanes × log): the path for a caller who
already knows the candidates (a database pre-filter) when they are few
against the postings a window search would walk. Scores are the same
f32 sums as search_with: contributions added lane by lane in query
order.