Rationalize redis sentinel configuration

We currently have redis-sentinel running on the same VMs as redis for the persistent (redis-XX) and sidekiq (redis-sidekiq-XX) cluster nodes, but on separate VMs (redis-cache-sentinel-XX) for the redis-cache cluster (redis-cache-XX).

This is a little confusing at times as an SRE (it requires a distinct cognitive effort when thinking about sentinel to remember which cluster is configured in which fashion).

A single sentinel cluster can manage multiple redis clusters, so we can either:

1. Move the cache sentinel onto the redis-cache nodes directly, to converge on a 'sentinel-next-to-redis' design, or
2. Migrate sentinel to a single dedicated cluster of 3 sentinel machines that monitors/manages all of the redis clusters.

This is not a high priority, as there is no known technical problem being caused with the current design.