For instance, Castor may say the request timed out, but that doesn't mean the request has been removed from the queue.
We need to issue an abort in all cases.
1. If indeed it has been cancelled, it is a noop
2. If it happens to be on disk already, it will stay already there for the next request
3. If it hasn't been cancelled, avoids piling up
This is basically a misunderstanding between storage and FTS.