Use Mapping instead of dict to fix downsteam type complaints.

With this change, downstream code can list just what types they might
actually provide, rather than all of the ones the method could accept.

python/lsst/daf/butler/_dataset_provenance.py

@@ -78,14 +78,14 @@ def add_input(self, ref: DatasetRef) -> None:
         self._uuids.add(ref.id)
         self.inputs.append(ref.to_simple())
 
-    def add_extra_provenance(self, dataset_id: uuid.UUID, extra: dict[str, _PROV_TYPES]) -> None:
+    def add_extra_provenance(self, dataset_id: uuid.UUID, extra: Mapping[str, _PROV_TYPES]) -> None:
         """Attach extra provenance to a specific dataset.
 
         Parameters
         ----------
         dataset_id : `uuid.UUID`
             The ID of the dataset to receive this provenance.
-        extra : `dict` [ `str`, `typing.Any` ]
+        extra : `~collections.abc.Mapping` [ `str`, `typing.Any` ]
             The extra provenance information as a dictionary. The values
             must be simple Python scalars or scalars that can be serialized
             by Pydantic and convert to a simple string value.