Add FEType::p_refinement

[roystgnr]

This isn't technically a complete solution to #4253, because it still won't let us use different p-refinement settings in the same VariableGroup, but IMHO it's good enough to count as a fix for that issue - it lets us add_variable() with p-refinement disabled, and the DofMap won't automatically group together variables with and without p-refinement, so we can easily have variables with the same order and same family and different p-refinement levels.

I'm starting this PR in draft status because I need to add more tests before we can consider merging it, but it's at least at the point where I want to see if it breaks any of the MOOSE should_p_refine() test coverage.

[roystgnr]

I swear I did compile that myself before pushing ... but I might have only compiled it in a --enable-everything build. Here's hoping I can fix the --disable-ifem case blind.