#1505 Make MoReFEM work with Petsc 3.12
Models that uses up non linear solve no longer works, and after discussion with Petsc people it's clear I made an assumption I shouldn't have: the output residual given to the SNESFunction is intended as a work variable and I shouldn't link it as closely to SystemRhs (it's likely they now swap the pointer in their internals, which causes the discrepancy in output I now have)
It appears as well that one of my sanity check was in fact faulty as it relied completely on the aforementioned assumption.