Currently resume tokens for resharding improvements work with aggregation only when the pipeline is otherwise empty aside from the initial document scan stage (this works because the pipeline executor is optimized out). We need to pass the resume token through the pipeline executor.