Allow p-refinement enable/disable per variable, not simply by group

[lindsayad]

For @YaqiWang and just better overall "moral correctness" (credit to @roystgnr)

[roystgnr]

As per ranting in idaholab/moose#31563 (comment) - it seems like the way to do this is to consider variables with enabled vs disabled p refinement to be of different FEType, just as we do now if they're of different families (even though e.g. every C0 family is the same at p=1 and every discontinuous family is the same at p=0) or of different orders.

I was previously thinking that what we needed to do here was to move p-refinement status from DofMap to Variable, but moving it to FEType seems to be both more "morally correct" and more useful (in the sense that it'll require fewer changes there) downstream.