Remove cuda from explicit environments.

We added these so that we could pip install tensorflow on Windows,
but since anaconda now has tf 2 we're back to using conda for tf.
Explicitly requiring cuda means we download them even when we
don't want gpu support--i.e., during CI testing.

environment.yml

@@ -5,8 +5,6 @@ dependencies:
 - attrs
 - jsonpickle=1.2
 - networkx
-- cudatoolkit=10.0.*
-- cudnn
 - tensorflow-gpu=2.0
 - scikit-learn
 - h5py