Implement a auto re-trying wrapper backend

[grote]

From personal observation issues like #675 are the most common cause of backup failures. While okhttp has auto-retrying on, it doesn't seem to kick in for situations like those (EOFException, SocketTimeoutException, etc.).

A solution could be do wrap existing backends in a RetryingBackend which delegates all calls to the underlying backend, however, if it receives a non-permanent exception, it retries the operation, potentially with some back-off. This way, the retrying logic would be hidden from the rest of the app to reduce complexity and improve test-ability.

For a start, we'd declare all exception permanent failures, except the ones we've seen to be transient in the wild.