Redesign monitoring internals

[davetron5000]

Treating the rake task as the public interface here, but this would break anyone using the stuff in lib/monitoring

• Created a real data type for the results of checks
• Removed the "per queue" notifier, which was tightly coupled to its checker. Now librato notifier can handle all of them
• This should make it easier to add more advanced checks/notifications.

To get confidence this won't break anything, I basically made the internal changes and kept running the monitoring integration test. Once I'd made all the changes without that breaking, I then updated the unit tests.

(The change in the monitoring integration test was just to add failure messages to the asserts. No actual changes were made)

This is in advance of another change I'm going to make to add stats around the class of failed jobs, so we can treat certain failures as more urgent than others without having to create a custom resque error handler. Wanted to get feedback on this first, which will make the overall change easier to understand (hopefully :)

[davetron5000]

This now captures the contract between checker and notifier. Before it was basically whatever and so it resulted in tight coupling between some checkers and notifiers.