Convergence criterion

[Ivar Stefansson]

The current default convergence criterion is based on the iteration solution update. I suggest we consider replacing it with a residual based one. This might wait for the planned AD parsing overhaul which hopefully allows for evaluation of the residual without re-computing the Jacobian.

[Eirik Keilegavlen]

Based on input from @jwboth, we should also consider monitoring the residual for individual equations, rather than only for the full system.

[Ivar Stefansson]

This is now facilitated. Let's make a decision on whether to make residual based convergence default, @keileg and @jwboth