Spike: Investigate duration of #realtime_changes endpoint

The EpicsController#realtime_changes endpoint receives a lot of traffic as it is polled once every 3 seconds for every open Epic.

A significant number of these take over 1s, the target duration for this endpoint. Looking at the spread of duration on downstream services for this subset, most of this time seems to be spent in Redis:

https://log.gprd.gitlab.net/goto/f95933d0-cd60-11ec-afaf-2bca15dfbf33

Drilling further down, it's specifically the Redis Cache:

https://log.gprd.gitlab.net/goto/75d88a50-cd61-11ec-b73f-692cc1ae8214

---

It's worth investigating if this can be improved, as it's presumably the same for issue and merge request realtime_changes and this would represent a lot of error budget spend and pressure on Redis.