Use tenacity instead of retrying
retrying is unmaintained and tenacity is its replacement. This also reworks the retrying logic slightly by adding exponential backoff, thus increasing the wait time for transient issues even more.
It should now retry for 5 times with 2min, 4min, 8min, 16min, 30min in-between attempts, covering an hour.
Clean up the CI suites while I am at it.