Add to checkmk the metric counting machines in `Error` state on auto-scaled runners managers

We have some servers that are managing the auto-scaled runners (for shared runners on GitLab.com and for GitLab CE/EE projects). We should count the number of machines in Error state and set a proper threshold that will notify us when the count of Error machines will be too big.

Me or @ayufan will update the Runners documentation to describe why that can happen and what to do in such situation.