Increase timeout to 80s. Backoff poll frequency. - #902

This is a minor improvement only.
Won't address the fails on cnds, maybe more luck on aov.

To be frank, I don't really like that we use polling here. But if it must be, maybe we can decide in advance what a reasonable amount of time for this operation would be like. And then we might use an approach such as this (taken from an earlier version of the entropy test):

def retry(func, exc_type, timeouts=(8, 7, 15, 10, 20, 30, 60)):
    if isinstance(exc_type, str):
        exc_type = exc_type.split(',')
    timeout_iter = iter(timeouts)
    # do an initial sleep because func is known fail at first anyway
    time.sleep(next(timeout_iter))
    retries = 0
    while True:
        try:
            func()
        except Exception as e:
            retries += 1
            timeout = next(timeout_iter, None)
            if timeout is None or e.__class__.__name__ not in exc_type:
                raise
            logger.debug(f"Initiating retry in {timeout} s due to {e!r} during {func!r}")
            time.sleep(timeout)
        else:
            break
    if retries:
        logger.debug(f"Operation {func!r} successful after {retries} retries")

It uses a sequence of times instead of a fixed function. Depending on what time we expect the backup to take, we could start with a high initial number, such as 30, and then continue with 15, 10, and then increase again, maybe 15, 20. In sum, this would be 90 seconds?

Well, it's hard to predict the time it will take.
As the storage system can do things like copy-on-write for backups, this may well be done within 2s (which is basically the call overhead).
If on the other hand, the backup cluster is separated from the main storage, it will have to copy over data. For 1GB, this could reasonably take up to 10s (100MB/s effective copy rate). Now in a busy system, there may be congestion or internal retries, so we should not give up too early.
So we could do some waiting, 2+3+5+10+15+25+40 = 100s and then consider it failed.