[feature] exponential backoff for retries

[hpgrahsl]

The current behaviour of the connector w.r.t retries is a pretty naive one. It would make a lot of sense to implement an exponential backoff mechanism as a more robust retry policy. Thus all errors for which it is reasonable to have retries in the first place should respect new retry settings that allow for non-linear timeouts, ideally including randomized jitter. Also this new approach to deal with timeouts on retries can/should completely replace what's currently there.