Introduce Rack middleware to discover CPU-intensive endpoints
We should introduce Prometheus metrics/histogram that would measure real-time spend when executing CPU thread, PROCESS_THREAD_CPU_TIME
and extend our existing metrics with it to ask for optimization and prioritization.
We need to do it for Unicorn and Puma.
Please check https://gitlab.com/gitlab-org/gitlab-ce/issues/64703 for more details, implementation proposal, and the discussion.