fix(discovery): list and watch loop without backoff when every watch gets 410 Gone #181

The Kubernetes provider can hammer the API server with LIST requests. After a successful re-list it resets the backoff counter and opens a new watch right away. If the server answers every new watch with 410 Gone at once, the provider repeats list and watch with no delay. The reset came with a09adab.

Fixed in #177. The counter is now reset only when the watch that ended in 410 was open for at least 30s; otherwise the provider waits the backoff.