PipelineTask preflight supports taking inputs from multiple collections, including specifying that certain DatasetTypes should be retrieved from specific collections.  Robert Lupton has calibration-production use cases for doing that with just a Butler, which, if implemented, might also lead to some simplifications elsewhere (e.g. using the QueryBuilder classes in more Registry methods).

It could possibly also be a first step towards a lazier, more Gen2-like approach to collection-chaining, which could be nice if we either find out that the new model doesn't support existing use cases in other ways of it leads to DB performance problems due to a massive number of tag records.