Retry with backoff for Gitlab API and git operations

[LinuxMercedes]

It seems that Gitlab can sometimes fall over under a barrage of requests. We've not had much luck getting Gitlab to be more reliable, so we should work around that in Assigner by retrying operations if they fail.

• Delineate between an actual failure and something worth retrying
• Offer some way for the user to specify the max number of retries
• Exponential backoff

[BehindTheBrain]

For someone who had this issue with assigner (but tested timeouts in custom git-python scripts rather than assigner), 1.5 to 2.5 seconds fixed delay seems to work most of the time for avoidance.