Use identity instead of hash for caching subfields in collect_subfields

This is closer to what GraphQL.js does, and performs much better:
benchmark_executing_introspection_query went from Mean 1.1484 to 766.0428

Author: ktosiek

src/graphql/execution/execute.py

@@ -204,9 +204,7 @@ def __init__(
         self.type_resolver = type_resolver  # type: ignore
         self.middleware_manager = middleware_manager
         self.errors = errors
-        self._subfields_cache: Dict[
-            Tuple[GraphQLObjectType, Tuple[FieldNode, ...]], Dict[str, List[FieldNode]]
-        ] = {}
+        self._subfields_cache: Dict[Tuple[int, int], Dict[str, List[FieldNode]]] = {}
 
     @classmethod
     def build(
@@ -998,7 +996,7 @@ def collect_subfields(
         subfields are not repeatedly calculated, which saves overhead when resolving
         lists of values.
         """
-        cache_key = return_type, tuple(field_nodes)
+        cache_key = id(return_type), id(field_nodes)
         sub_field_nodes = self._subfields_cache.get(cache_key)
         if sub_field_nodes is None:
             sub_field_nodes = {}