Related to #235. This changes the how the recovery endpoint reports success. Instead of synchronously returning when recovery is successful, it will return success when the recovery attempt has begun. A separate endpoint will exist to allow the state to be polled.