Use custom metric for sidekiq HPA
Summary
During some load testing of sidekiq I was doing today as part of https://gitlab.com/charts/gitlab/issues/809, I saw that while sidekiq tasks can take a lot of cpu, the majority, (which end up being pipeline related) end up mostly not occupying a lot of cpu, but rather just maxing out the available workers.
For out HPA, the targetCPUValue is properly causing sidekiq to scale up under pipeline load, but it could like do so more effectively if we also had it scale up when over 70% of the sidekiq workers were being occupied.