[batch] Fix batch local mode in jupyter notebooks (#13614)

The `LocalBackend` recently was changed to include a call to
`async_to_blocking` but does not apply `nest_asyncio`, so using the
local backend in a jupyter notebook breaks. I lifted the application of
`nest_asyncio` from the `ServiceBackend` to the super class. Should we
add a test that runs in its own python session to make sure that this is
caught?

hail/python/hail/docs/index.rst

@@ -38,6 +38,3 @@ Indices and tables
 * :ref:`genindex`
 
 If you would like to refer to our Hail v0.1 (deprecated) docs, please view `Hail 0.1 docs </docs/0.1/index.html>`_
-
-
-

hail/python/hailtop/batch/backend.py

@@ -48,11 +48,12 @@ class Backend(abc.ABC, Generic[RunningBatchType]):
     """
     Abstract class for backends.
     """
-
     _closed = False
 
     def __init__(self):
         self._requester_pays_fses: Dict[GCSRequesterPaysConfiguration, RouterAsyncFS] = {}
+        import nest_asyncio  # pylint: disable=import-outside-toplevel
+        nest_asyncio.apply()
 
     def requester_pays_fs(self, requester_pays_config: GCSRequesterPaysConfiguration) -> RouterAsyncFS:
         try:
@@ -486,11 +487,7 @@ def __init__(
         gcs_requester_pays_configuration: Optional[GCSRequesterPaysConfiguration] = None,
         gcs_bucket_allow_list: Optional[List[str]] = None,
     ):
-
-        import nest_asyncio  # pylint: disable=import-outside-toplevel
-
         super().__init__()
-        nest_asyncio.apply()
 
         if len(args) > 2:
             raise TypeError(f'ServiceBackend() takes 2 positional arguments but {len(args)} were given')