To support subpipelines in viewless timeseries collections we need to have collection catalog data in makePipeline(). This way we can call the timeseries translation logic after we parse the pipeline but before we optimize the pipeline. However, for queries on sharded clusters we must acquire collection catalog data using a RoutingContext inside a single CollectionRouter loop. That means we must call makePipeline and preparePipelineForExecution inside a single CollectionRouter loop. This involved a large refactor of $lookup and requires new process interface functions.