This change fixes an inconsistency where the mongos getMore reply for an invalidated change stream omitted cursor.invalidated: true, while the equivalent reply from mongod (first batch and getMore) and the mongos aggregate first-batch reply already set it.

A change stream cursor on a sharded cluster is invalidated by a collection drop/rename or database drop, and the router cursor throws ChangeStreamInvalidated during getMore. The closing reply (already id: 0 with the PBRT set to the invalidate token) now also reports cursor.invalidated: true.

Before: cursor: { id: 0, ns, nextBatch, postBatchResumeToken }
After:  cursor: { id: 0, ns, nextBatch, postBatchResumeToken, invalidated: true }

CursorResponse only serializes the field when true, so every other reply (e.g. normal batches, ordinary exhaustion, non-change-stream cursors) remains unchanged.