At this moment mongo parses MQL query twice: using lite parser and "heavy" one (::createFromBSON). Those a two distinct stages with almost no shared information. This leads to complexity (duplication of the code). Lite parsing lacks certain information to make informed decisions.

In this particular case, it is necessary for lite parsing to have access to sources constraints in order to reject or accept an embedded pipeline on $lookup stage. The current check is based on the name of the first stage in the pipeline which is unnecessarily specific.