Also kill the kink logic and thus minimisation2

Kill tests that use the old interface

Also fix the order a-mu-V0-stress-b-state-L everywhere

We use lumping for the nonlinearity, thus we only have diagonal entries

Drop members a, V_0, K
Move a into coefficientProduct
Move V_0 into V_m

Add comments that show what we're computing

Do not re-create overallSolver for every timestep

At least for the tangential descent, a gradient of magnitude 1e-16
causes problems. Test included