For some problems, especially for linear problems and zero initial value problems, we want to be able to set a constant epsilon per primary variable for numeric differentiation.
This introduces a new epsilon class that reads two parameters from the input file
the magnitude is a vector with the number of privars as size, base eps (optional) is a scalar (default 1e-10). If nothing is set it default to the old behaviour (trying to estimate the magnitude from the primary variable at the d.o.f the residual is assembled at (which leads to different epsilons in different cells).
For multidomain problems those parameters can be set per TypeTag / ModelParameterGroup.