Collect error rate data for tools

It would nice to identify tools that are probably bugged: we can (somehow) look into the job exit codes on different usegalaxy.* servers, and tag tools with a "to fix" label when we see most of the jobs are failing.

I guess a tool can fail on a specific server and work fine on another one, not sure how to handle that (only tag when it fails everywhere? warn specific server admin where it fails?)

I recall @neoformit had some magic sql queries to identify problematic tools, maybe it could be used somehow?