[Time series Gap-fill V2] Server side gap-filling of time series data

We recently introduced gapfill() to interpolate and fill gaps in a time series dataset. In the current solution, the data is moved and gapfilling is performed in the broker. This puts lots of stress on the broker as all of the raw records to be transferred from server to broker.

This brings in a lot of limitations and doesn't work for large dataset with larger date range. In V2, let us revisit the implementation to push down the gapfill and aggregations to server side by leveraging data locality and make this feature scale better.

cc @weixiangsun @Jackie-Jiang

Start by locating the existing gapfill() implementation and tracing how raw records and aggregations move through the broker. Then identify the server-side query path and the tests covering gap-filling. Done means gap-filling and aggregations execute server-side, reduce broker data transfer, and work for large datasets and date ranges.

Written by the indexing model from the issue text.