Sub-cluster support for environments

With the "gpu" environment we're mixing up the dimensions of prod/staging with "function". That works in our favour at the moment as compiler discovery means we need the "gpu" cluster up while we "discover" compilers. However, it's inelegant for management, we need to remember to update "gpu" separately from prod, and there's not a way to test changes in the one gpu cluster. Of course we can make a "staging.gpu" environment...but...

Maybe we should have the concept of "cluster" underneath environment? That is, tooling should support understanding that an env contains multiple clusters/ASGs? That opens the doors to other things like "ARM cluster" and/or "separate execution cluster" or "compiler unpacking cluster" (for manyclangs) or whatever.