SSH: Gracefully recover after partial git bundle upload

[ammario]

See #2364 for background. Regardless of the underlying transport bugs we should recover better, as that class of issue could recur due to run of the mill network flakiness.

[ethanndickson]

The current behaviour when the git bundle transfer is cut off is that we retry up to three times. The linked issue was persistent so those retries weren't helpful. Maybe indefinite retries with a backoff is necessary? What other sort of recovery did you have in mind?