Before, it was in the LoopSolver class, which derives directly
from IterativeSolver.

I think the original idea was that anything that is an iterative
solver in the mathematical sense should derive from IterativeSolver.
This includes things like the QuadraticIPOptSolver, which is the
implementation of an iterative algorithm, inherits from Iterative-
Solver, but where you cannot get the individual iteration step
as a dune-solvers object.

I started to think that this is the wrong approach.  The inheritance
should be governed less by what an object 'is' in the mathematical
sense, but rather by the kind of API it provides.  And I would
expect an IterativeSolver to provide an IterationStep.