createCanonicalQuery currently modifies its source pipeline before it creates a CanonicalQuery, so in case of a failure we can end up with a modified pipeline. This is fine in cases when failure of creating a CanonicalQuery object means failure of the query. However, in case of the join optimizer it only means that we should bail out from creating JoinGraph. To work around this issue we create a $lookup stage in the join optimizer when we try to create a CQ from it, we should consider if can avoid creating this copy.