Prioritize registered datasets over datasets in the datasets folder

If users have their own datasets, they are not in the datasets folder, but they will be part of the legacy dataset provider.

PiperOrigin-RevId: 686488188

Author: tomvdw

tensorflow_datasets/core/registered.py

@@ -300,8 +300,12 @@ def get_builder_cls(self, name: str) -> Type[RegisteredDataset]:
     ...
 
 
-class LegacyDatasetBuilderProvider(DatasetBuilderProvider):
-  """Provider of dataset builders that are defined in the legacy codebase."""
+class RegisteredDatasetBuilderProvider(DatasetBuilderProvider):
+  """Provider of dataset builders that are defined by builder classes.
+
+  If there's a builder class, then it will be registered because it's a subclass
+  of RegisteredDataset.
+  """
 
   def has_dataset(self, name: str) -> bool:
     if name not in _DATASET_REGISTRY:
@@ -373,8 +377,8 @@ def get_builder_cls(self, name: str) -> Type[RegisteredDataset]:
 
 def _get_inital_providers() -> list[DatasetBuilderProvider]:
   return [
+      RegisteredDatasetBuilderProvider(),
       SourceDirDatasetBuilderProvider(constants.DATASETS_TFDS_SRC_DIR),
-      LegacyDatasetBuilderProvider(),
   ]