Core service scaling: separate RQ workers

[olevski]

From #3094

Dealing with the RQ workers:

• Should we separate out the normal workers with their own ephemeral cache for background jobs and a common queue?
• Some queues need to be local to each service (those clean up the local cache after ~30 minutes and also clean up the uploaded file cache)
• if we decide to do the above in a follow-up, we need at least separate redis queues per core-service instance so they don't interfere with each other

[Panaetius]

check #3100 first