Fix overly strict check in local_pow_specialize

Author: ricardoV94

pytensor/tensor/rewriting/math.py

@@ -2072,11 +2072,10 @@ def local_pow_specialize(fgraph, node):
             if np.all(y == -2):
                 rval = [reciprocal(sqr(xsym))]
             if rval:
+                if not rval[0].type.broadcastable == node.outputs[0].type.broadcastable:
+                    return None
                 rval[0] = cast(rval[0], odtype)
-                assert rval[0].type.is_super(node.outputs[0].type), (
-                    rval[0].type,
-                    node.outputs[0].type,
-                )
+                assert rval[0].type.dtype == node.outputs[0].type.dtype
                 return rval
     else:
         return False