Now that we allow $projections (and soon renames) in subpipelines, we should allow queries like this to participate in join optimization:

{$lookup: {from: "A", let: {lf: "$a"}, as: "fromA", pipeline: [
   {$project: {renamedField: "$b"}},
   {$match: {$expr: {$eq: ["$$lf", "$b"]}}}
]}},
{$unwind: "$fromA"}

It shouldn't matter if the $match is at the start or end of the pipeline, since we will keep the $project in our CQ & encode the edge into the join graph (i.e. once this goes to SBE, the edge is "executed" after the $project even if you swap the $match/$project, and this is correct regardless).