Cache pydantic model fields for faster decoding

In `timeit` tests, 10.000 calls to `ModelField()` could take up to half
a second on my Macbook Pro M1, depending on the type annotation used.
Given that the method is called for every cache hit, this can really add
up. The number of different return types for endpoints is very much
finite however, so caching is a definite win here.

fastapi_cache/coder.py

@@ -3,7 +3,7 @@ import datetime
 import json
 import pickle  # nosec:B403
 from decimal import Decimal
-from typing import Any, Callable, TypeVar, overload
+from typing import Any, Callable, ClassVar, Dict, TypeVar, overload
 
 import pendulum
 from fastapi.encoders import jsonable_encoder
@@ -53,6 +53,13 @@ class Coder:
     def decode(cls, value: str) -> Any:
         raise NotImplementedError
 
+    # (Shared) cache for endpoint return types to Pydantic model fields.
+    # Note that subclasses share this cache! If a subclass overrides the
+    # decode_as_type method and then stores a different kind of field for a
+    # given type, do make sure that the subclass provides its own class
+    # attribute for this cache.
+    _type_field_cache: ClassVar[Dict[Any, fields.ModelField]] = {}
+
     @overload
     @classmethod
     def decode_as_type(cls, value: str, type_: _T) -> _T:
@@ -72,9 +79,12 @@ class Coder:
         """
         result = cls.decode(value)
         if type_ is not None:
-            field = fields.ModelField(
-                name="body", type_=type_, class_validators=None, model_config=BaseConfig
-            )
+            try:
+                field = cls._type_field_cache[type_]
+            except KeyError:
+                field = cls._type_field_cache[type_] = fields.ModelField(
+                    name="body", type_=type_, class_validators=None, model_config=BaseConfig
+                )
             result, errors = field.validate(result, {}, loc=())
             if errors is not None:
                 if not isinstance(errors, list):