Expand Retry Logic

[jpoehnelt]

It would be nice if the pusher retried on 504s and other similar responses that occur when backends are having issues. For example, today: https://status.cloud.google.com/incident/cloud-dev-tools/18003#18003003

[drigz]

+1, I ran into this when pushing multiple images in parallel to a newly created projects (with rules_docker's container_push). The first request succeeds, but the next 10 or so fail with a 503.

[fahhem]

We've (not Google) done this by overriding/extending the ShouldRetry function in the RetryTransport class, which was nicely setup to allow that. Is that useful?