[BUG] get_combined_model() for dask knn returns None

get_combined_model() for dask knn returns None

This function works for dask rf, but not knn.

e.g. this little snippet returns None after model was fitted fine:

from cuml.dask.neighbors import KNeighborsRegressor as KNeighborsDaskRegressor
from cuml.dask.neighbors import KNeighborsClassifier as KNeighborsDaskClassifier
if isinstance(model, (KNeighborsDaskRegressor, KNeighborsDaskClassifier)):
    model = model.get_combined_model()

Using rapids 21.08, might be fixed in later versions?

Start by reproducing the provided snippet with KNeighborsDaskRegressor and KNeighborsDaskClassifier, using the reported RAPIDS 21.08 context. Compare get_combined_model() behavior with dask random forest and verify that the fitted dask KNN models return a combined model instead of None.