[Data] Deprecate read_parquet_bulk (ray-project#48691)

Users (including Ray Data developers!) are often confused about how to
choose between `read_parquet` and `read_parquet_bulk`. To avoid
confusion, this PR deprecates `read_parquet_bulk`.

Signed-off-by: Balaji Veeramani <bveeramani@berkeley.edu>
Signed-off-by: hjiang <dentinyhao@gmail.com>

python/ray/data/read_api.py

@@ -83,7 +83,7 @@
 from ray.data.datasource.parquet_meta_provider import ParquetMetadataProvider
 from ray.data.datasource.partitioning import Partitioning
 from ray.types import ObjectRef
-from ray.util.annotations import DeveloperAPI, PublicAPI
+from ray.util.annotations import Deprecated, DeveloperAPI, PublicAPI
 from ray.util.scheduling_strategies import NodeAffinitySchedulingStrategy
 
 if TYPE_CHECKING:
@@ -923,7 +923,7 @@ class   string
     )
 
 
-@PublicAPI
+@Deprecated
 def read_parquet_bulk(
     paths: Union[str, List[str]],
     *,
@@ -1023,6 +1023,12 @@ def read_parquet_bulk(
     Returns:
        :class:`~ray.data.Dataset` producing records read from the specified paths.
     """
+    warnings.warn(
+        "`read_parquet_bulk` is deprecated and will be removed after May 2025. Use "
+        "`read_parquet` instead.",
+        DeprecationWarning,
+    )
+
     if meta_provider is None:
         meta_provider = FastFileMetadataProvider()
     read_table_args = _resolve_parquet_args(