Recovery from failures

[mikedilger]

When we are unable to do something at a relay, because it is down or whatever, we could save what we were trying to do in the database, and try again later, exponentially backing off, and eventually giving up.

This would require a new database to keep the ToMinionMessage objects (or similar) along with a timestamp when we last tried, and how big the exponential backoff currently is.

This would also require a background task that periodically checks these to see if any need to be tried again, or given up upon.