#1 Streaming paging
Currently the tableflow mechanism works in such a way that the entire dataset is retrieved from the datasource. This is among others required for the ranged pager, to learn about the range of values in each column and learn about the number of rows available.
For some situations however, loading the entire dataset into memory is less than required. There are various use cases that call for streaming paging: pages are requested from the datasource on demand.
There are certain implications if that was implemented, for one the pager would not know about ranges anymore. Furthermore, various operations that are currently executed at the caching site (for instance sorting and filtering) will have to be re-implemented to be performed at the (remote) data source.
