NC | GPFS | Threaded Multiple Object Delete fix

[romayalon]

Explain the changes

1. When calling delete multiple objects on the same key & version concurrently, the version might move/delete, on GPFS we saw that _open_files_gpfs() failed on ENOENT in that case, therefore added a retry when got ENOENT, next iteration we will return if the version was deleted or get the new location of the version.

Issues: Fixed #xxx / Gap #xxx

1. Fixed NSFS | S3 | Versioning: Threaded delete of multiple objects does not delete #8377
2. Gap - Enable s3tests_boto3/functional/test_s3.py::test_versioning_concurrent_multi_object_delete, currently it's not working because they call put bucket twice and the test fails on BucketAlreadyExists, the test I added dooes the same as the ceph test but in general we want to remove it from the pending list. Opened a bug - test_versioning_concurrent_multi_object_delete creates the bucket twice and fails on BucketAlreadyExists ceph/s3-tests#588.

Testing Instructions:

1. Run sudo jest -t 'concurrent multi object delete' on GPFS cluster or run s3tests_boto3/functional/test_s3.py::test_versioning_concurrent_multi_object_delete while fixing manually the bug on s3-tests code - see the suggested fix in test_versioning_concurrent_multi_object_delete creates the bucket twice and fails on BucketAlreadyExists ceph/s3-tests#588.

• Doc added/updated
• Tests added