Remove failed deployments earlier in the filtering
Previously, we only fetched the two latest deployments, so if the first
was running and the second failed, we'd end up with only the running
deployment, meaning auto_deploy status never showed the current
version for that environment.
Now we fetch 10 and reject failed deploys before we check if there's nothing upcoming.
This means we should always get back the latest successful deploy, plus an upcoming deploy if there is one, but never get back only the running deploy.