Improve alerting for CI runners outages

Some ideas:

• Alert on the rate of failed DO creates
• Alert if the number of running jobs is low, especially if the number of pending jobs is high, compared on a per-type basis (so GCE jobs won’t obscure DO failures)

In general, we should be able to alert when our provider is having a failure so we can react quicker to the situation by enabling the secondary (more expensive) shared runners.

@tmaczukin do we have enough metrics to alert on rate of failures vs rate of successes? (we are doing this for the web front end and it works really well)