This change adds an SBE optimization in the collection-scan stage builder: a scan only materializes the full document (the "result object" / recordSlot) when a downstream consumer needs it (`reqs.hasResult()`) or the scan's own filter needs the whole document. Otherwise it produces only the referenced top-level fields plus the record id.

Key changes:

• `SbBuilder::makeScan()` gained an `includeResultSlot` flag and now returns a `boost::optional<SbSlot>` result slot.
• `generateGenericCollScan()` / `generateClusteredCollScan()` compute `includeResultSlot` and skip materialization when possible (generic path extracts the filter's top-level fields; clustered path keeps the result object whenever there's a filter).
• `buildCollScan()` passes `reqs.hasResult()` down.