Improve monitoring to catch memory growth problems earlier

[makortel]

Would like to catch situations like linear growth, spikes at the end of the job, early much before they start to cause problems (i.e. job failures) in production

Could be something like

• SimpleMemoryCheck monitors RSS, reports some growth metrics in the framework job report
• WM propagates these metrics to monitoring
• monitoring raises alerts if certain patterns start to occur
• operator responding to alert opens CMSSW GitHub issue

[makortel]

This campaign has effectively been superseded by #670