Make the json logging more suitable for ELK indexing

This issue is imported from pivotal - Originaly created at Dec 9, 2020 by Bart Geesink

Currently we have json logging, which enables us to send the errors to ELK. We use ELK to view what's happening when you release a new version, or to do analysis of certain incidents.
It is now difficult to do analysis, because the "message" does not contain enough information (only stuff like "error occured" or "Servlet.service() for servlet [dispatcherServlet] in context with path [] threw exception ". The other field with information is the stack trace itself.

I'd like to be able to get a field that I can use to aggregate and count what's happening.

So the field "message" would contain keywords, like
"missing_refresh_token"
"code_used_twice"
"no_access_token_found"
etc

This would allow us to see trends, and check for specific errors in certain RP's

No file or test is named. Start by locating the JSON logging configuration and the code paths that emit the example errors, then assess how the message field is consumed by ELK. Done means the relevant errors expose consistent, aggregatable message keywords without losing the existing stack-trace information.