Fix structured kernel codegen (#49244)

Summary:
Pull Request resolved: #49244

see https://fb.quip.com/ceEdANd5iVsO

RegisterMkldnnCPU kernels incorrectly used makeUnboxedOnly() calls to register add_.Tensor kernels. This is because the codegen incorrectly thought they're not c10-full.
This PR fixes that.
ghstack-source-id: 118411117

Test Plan: After this PR, RegisterMkldnnCPU doesn't contain the makeUnboxedOnly() calls anymore.

Reviewed By: ezyang

Differential Revision: D25500246

fbshipit-source-id: 8a8c2be9c4f4a5ce7eaae94257c2f8cbd176e92e

tools/codegen/gen.py

@@ -471,6 +471,7 @@ def gen_one(f: NativeFunction) -> Optional[str]:
 
         return list(mapMaybe(gen_one, g.functions()))
 
+    @method_with_native_function
     def gen_unstructured(self, f: NativeFunction) -> Optional[str]:
         # for mypy type refinement; would be fixed by TODO on target
         assert self.target is not Target.DECLARATION