This is useful when the user (a puppet provider) is expecting a error.
With this it doesn't have to wait for the long retry cycle to have its
error back. This replace the awkward mechanism in
self.request_without_retry which was dynamically mangling
request_timeout and disabled even legit retry.
Examples of such function are self.fetch_user, self.fetch_project in the
keystone provider.
Change-Id: I589da0100ad3ccf3abf17ac8b26c827793ace484
Related-Bug: 1597357