BMC install is not resilient against network glitches

[cybertron]

It has been found that in clouds where API calls are not especially robust (possibly due to network issues or API overloading), the BMC install will often fail because it needs to make a number of API calls and has no retry mechanism if one of them fails. It would be nice to add a retry function so those calls can potentially still succeed if they hit a transient issue.

Note that this is somewhat mitigated by a36b6b7 because that allows BMC failures to be found much earlier. The OVB stack will correctly fail now if the BMC fails, instead of requiring the deployment to get all the way through to attempting an IPMI call before the problem is discovered.