We anticipate many customers to expect this behavior out of the box, as it is how other vector db's often execute hybrid search. Considering $rankFusion as a special case of agg, and executing search subpipelines as a special case of $rankFusion, what options do we have to parallelize their execution?