ProjectionPathASTNode getChild performs a linear search through a vector of field names to find the correct child. Because of this search, we see n^2 behavior on the number of fields in the $project. As we add children to this node in addNodeAtPathHelper, we also call getChild to search through everything we've added so far.

ProjectionPathASTNode could use a map instead of a vector to avoid this behavior.

This ticket should also involve performance testing with a map. If we see a regression, we could consider using a map when there are >50 (or some N) elements.