Add py-torch (in some non-cuda version)

Egidio Falotico mentioned NRP has problems integrating py-torch "into
the platform". Let's see what happens…

spack.yaml

@@ -29,4 +29,6 @@ spack:
     #- pynn-brainscales@1.0-rc2 ^log4cxx@0.10.0 ^googletest@1.11.0:+gmock %gcc@10.3.0
     # demo for codejam12
     - funniest1022 %gcc@10.3.0
+    # NRP
+    - py-torch~mkldnn~cuda~cudnn %gcc@10.3.0
   concretization: together