Use persistent Redis cluster for Workhorse pub/sub notifications

Previously, in Omnibus, Workhorse expected to listen via the Redis shared state cluster for the workhorse:notifications publish/subscribe channel, but the Rails code was using the Sidekiq queue cluster for this. To fix this inconsistency, we make the Rails code use the persistent cluster, since we don't want Workhorse to be looking at anything Sidekiq-related.

Originated from gitlab-com/gl-infra/production#963 (closed).