When the invalidated_cursors.js workload is run in the concurrency_replication_causal_consistency suite, the $currentOp aggregation is sent to the pinned secondary, but the killOp command is sent to the primary. This can cause the test to fail due to an operation unexpectedly failing on the primary. We should ensure that the killOp command is sent to the same node as the $currentOp aggregation.

This probably also applies to the killCursors command in the same test, though that is less likely to collide across two nodes, since it is chosen randomly, unlike opIds, which are chosen incrementally.

This is the only FSM workload that is run in the causal consistency suites and uses killOp. We may experience a similar issue with killCursors in kill_multicollection_aggregation.js.