Collections with primary key can support more efficient cursors

From @quodlibetor: When collections have primary keys, we should be able to perform key-wise actions through a cursor by traversing new-to-old batches and looking for the first positive accumulation. Once identified, this should be the only record, under the hypothesis that each prior collection accumulated to something non-negative. This seems like an eminently plausible case, and the optimization of avoiding seeks into the large batches is appealing.

The issue names no file, test, or entry point. Start by locating the collection cursor implementation and existing key-wise actions, then determine how cursor traversal handles accumulated batches. Done means the proposed primary-key case is covered by tests and avoids unnecessary seeks without changing results.