Fix another potential reuse of NoneConst between fgraph and doc how to use it.

theano/tensor/subtensor.py

@@ -1704,7 +1704,7 @@ advanced_inc_subtensor1 = AdvancedIncSubtensor1()
 
 def as_index_variable(idx):
     if idx is None:
-        return NoneConst
+        return NoneConst.clone()
     if isinstance(idx, slice):
         return make_slice(idx)
     idx = theano.tensor.as_tensor_variable(idx)

theano/tensor/type_other.py

@@ -66,4 +66,7 @@ class NoneTypeT(Type):
     def __str__(self):
         return "None"
 
+# This is a variable instance. It can be used only once per fgraph.
+# So use NoneConst.clone() before using it in a Theano graph.
+# Use NoneConst.equal(x) to check if two variable are NoneConst.
 NoneConst = Constant(NoneTypeT(), None, name='None')