* A lot of refactoring the the groupby. Mainly to include both distinct and null-grouping

* Test for non-dask aggregations

* All NaN data needs to go into the same partition (otherwise we can not sort)

* Fix compatibility with SQL on null-joins

* Distinct is not needed, as it is optimized away from Calcite

* Implement is not distinct

* Describe new limitations and remove old ones

* Added compatibility test from fugue

* Added a test for sorting with multiple partitions and NaNs

* Stylefix