Move the `retry` module from `client-base`

[llucax]

In the future, when we move run_forever() it might be nice to be able to pass a retry strategy to it. A good reason to move the retry module from client-base to this repo.

Originally posted by @llucax in frequenz-floss/frequenz-sdk-python#1081 (comment)

[llucax]

• Improve resiliency of long-running async tasks frequenz-sdk-python#1081 (comment)

  ExponentialBackoff is hard to do with generic functions, because it needs to know if a previous attempt succeeded, so that it can reset its backoff interval.

  This is easy to do with streams, we just reset it after every incoming message. But for functions, we'd need something like a timer to reset the backoff interval, if the function hasn't failed for a certain interval, then we say it succeeded, for example.

• Improve resiliency of long-running async tasks frequenz-sdk-python#1081 (comment)

  Yeah, maybe we can have a strategy that actually takes into account for how long the task ran, so if last time it ran for a long time it waits very little to restart and if it failed immediately it waits quite a long time to restart 🤔