Timeseries rewrites occur here when we're in the midst of resolving a view definition into the full aggregation pipeline. With query stats, we need to handle parsing the view pipeline and user pipeline separately in run_aggregate. That is impossible for timeseries queries since these rewrites are only done over the full resolved pipeline, so viewPipeline + userPipeline is not equivalent to the full pipeline.

We should consider moving some of these rewrites into the DocumentSource structure somewhere after parsing but before optimization; rewrites that don't re-order pipeline stages and just modify fields internal to a stage could also be moved into resolveView.