RFC: Global Variables in MLIR

Sounds good. Any more comments here before I revise the proposal to address some of the comments above?

As for the allocation modeling, do folks think we need to model this as a allocation of a new resource type or not model this as an allocation at all?