Extend Retry logic to address OSErrors.

I've encountered a case where large bulk deletions throws an `OSError`. The GCS documentation points out that this is a general problem (https://cloud.google.com/storage/docs/deleting-objects#delete-objects-in-bulk). I did not expect, however, that this would throw an OSError instead of a 500 error.

In any case, this PR extends `gcsfs`'s retry logic to include internal errors that throw `OSError`s. My hope is that this fixes an associated issue in Pangeo Forge Recipes (pangeo-forge/pangeo-forge-recipes#406).

gcsfs/retry.py

@@ -48,6 +48,7 @@ class ChecksumError(Exception):
     google.auth.exceptions.RefreshError,
     aiohttp.client_exceptions.ClientError,
     ChecksumError,
+    OSError,
 )
 
 
@@ -119,6 +120,7 @@ async def retry_request(func, retries=6, *args, **kwargs):
             google.auth.exceptions.GoogleAuthError,
             ChecksumError,
             aiohttp.client_exceptions.ClientError,
+            OSError,
         ) as e:
             if (
                 isinstance(e, HttpError)