Handle client timeout when running CompleteMultipartUpload

[JustinKyleJames]

When CompleteMultipartUpload is executed on a very large file, the client may timeout and retry resulting in some confusing errors.

Here is the flow:

1. Client calls CompleteMultipartUpload.
2. The part files are streamed from the S3 API to iRODS in parallel.
3. The client is not sending or receiving data so it can have a socket timeout.
4. When it times out, the client may retry the CompleteMultipartUpload.
5. On the second call to CompleteMultipartUpload, the S3 API will attempt to open the data object. This will fail because the data object is in intermediate state as the previous uploads are still executing.
6. The client reports an internal server error.
7. The S3 API reports an error like the following: "Failed open data stream to iRODS - path=/tempZone/home/alice/alice-bucket/1Gfile".
8. iRODS reports an error about a missing replica token.

[JustinKyleJames]

Among the considerations, we might should prevent multiple simultaneous calls to CompleteMultipartUpload with the same upload id. That way we can send a more useful error message to the client.