Extractor document nodes need to be in the tree before processing them

Otherwise references to parent nodes don't work, which we need e.g. for
context time propagation.