Why only a single Device?

[lindsayad]

I know there are implementation reasons why there should only be a single Device. We ran into this in idaholab/moose#31037. However, per that issue I feel like it's reasonable in a multiphysics simulation for different application types (perhaps one a 1D system code and another a 3D high-fidelity CFD application) to want to execute on different (hardware) devices. Is this something that might be supported in the future? Or can this concept be supported by a single Device instance? I know that Device can be configured with multiple backends but my understanding per the documentation is that the priority list for the configured backends is fixed and so any given application will end up selecting the same backend for a given environment.