Allow BatchEncoding to be initialized empty. (#4316)

* Allow BatchEncoding to be initialized empty.

This is required by recent changes introduced in TF 2.2.

* Attempt to unpin Tensorflow to 2.2 with the previous commit.

src/transformers/tokenization_utils.py

@@ -173,7 +173,11 @@ class BatchEncoding(UserDict):
 
     """
 
-    def __init__(self, data: Dict[str, Any], encoding: Optional[Union[EncodingFast, Sequence[EncodingFast]]] = None):
+    def __init__(
+        self,
+        data: Optional[Dict[str, Any]] = None,
+        encoding: Optional[Union[EncodingFast, Sequence[EncodingFast]]] = None,
+    ):
         super().__init__(data)
 
         if isinstance(encoding, EncodingFast):