Is Lemmy made with horizontal scaling (a.k.a. launching more instances and have a load balancer proxy the requests to the various instances) in mind? It could help larger instances like lemmy.ml managing the load better rather than just putting it on a beefier machine.