FIX, ADD: kwds catchall for Negation node

Passing kwds through a negation (as in the learning case) breaks inference - adding a catchall so that the node can ignore unused inputs
solves IBM#26

May need a better mechanism for handling the abstraction of `traverse_execute` but this should suffice for now.

lnn/symbolic/logic.py

@@ -1312,7 +1312,7 @@ def upward(self, **kwds) -> torch.Tensor:
             None, _utils.negate_bounds(self.operand.get_facts(*groundings))
         )
 
-    def downward(self) -> torch.Tensor:
+    def downward(self, **kwds) -> torch.Tensor:
         if self.propositional:
             groundings = {None}
         else: