Change distributed planner to use logical plan rather than physical plan

I intend to change the distributed planner to work with logical plans rather than physical plans. This means that query stages will use logical plans, and the workers can run those plans on execution engines other than DataFusion.

In https://github.com/apache/arrow-datafusion-python/pull/190 we are adding support for running DataFusion logical plans on Polars and Pandas, for example. 
