[POC] Export Sidekiq metrics from separate server process

Goal

Implement a POC demonstrating that the approach outlined in &6409 (closed) works.

Approach

• Make SidekiqExporter (and its dependencies) available to Sidekiq::CLI (it runs outside the Rails context currently)
  • This is the key challenge here, since it accesses Settings and other things specific to the Rails app
• Make sidekiq-cluster process spawn a metrics server process and bind to a different port (e.g. 3809; 3807 is where sidekiq procs currently attempt to bind)

Outcome

• Existing worker(s) continue to export metrics via 3807 to allow for a smooth rollout
• Server process exports the same metrics via a different port (set from env variable)
• Metrics must be identical, processes must not interfere with one another
  • Specifically, workers never remove anything from the shared metrics directory; only the server process does. This will address the problem where metrics disappear because restarting workers attempt to wipe data from it.

This should lay the foundation for a clean cut-over later.

Notes / Assumptions / Questions

• All environments launch sidekiq-cluster rather than individual SK processes
  • Should be the case as of recently so that we either go through bin/background_jobs (GCK/GDK, from source) or through a service wrapper (Omnibus, CNG via gitlab-org/build/CNG!819 (merged))
• How will clean-up logic be handled? This was one of the points of friction
  • By moving all clean-up logic to the parent process, the problem of workers deleting each other's metrics should be solved because:
    • The server process starts before any child processes (i.e. workers) are spawned; each worker can therefore assume to operate on a clean directory
    • As soon as any child process dies, all children will be terminated and the parent process quits. Restarting it is therefore equivalent to a cold start and will also wipe all metrics.
    • More context here: #27818 (comment 251044947)