Wrong order of routines before nonlinear iteration for Darcy flow

[vlipovac]

In the SinglePhaseFlow setup there is an inconsistency in the order of updated and stored values of discretizations and permeabilities.

Looking at the method before_nonlinear_iteration we have the following order code executed:

1. SinglePhaseSolutionStrategy.before_nonlinear_iteration
   a. Updates stored flux values <---
   b. Super call to base SolutionStrategy
2. SolutionStrategy.before_nonlinear_iteration
   a. Calls set_discretization_parameters
3. SinglePhaseSolutionStrategy.set_discretization_parameters
   a. Updates stored permeabilities in a SecondOrderTensor <---
4. Back to SolutionStrategy.before_nonlinear_iterationan
   a. Calls rediscretize (which rediscretizes Upwinding from SinglePhaseFlowSolutionStrategy)

The two lines with arrows show that the flux values for Upwinding are evaluated and stored before permabilities are updated.

Now, for setups without non-linearities in the permeability tensor, this does not matter.
But for setups which include advanced mechanical effects, or setups which use a pressure equation with mobility terms in the tensor (frational flow in CF framework), this does matter.

The order of execution needs to be revisited, or possibly some methods need to be split to make a correct order possible given PorePy's mixin approach.

TODO once solved
Adapt SolutionStrategyNonlinearMPFA which was merged in #1236 .

[vlipovac]

Proposed solution:

This issue is critical for compositional flow and (thermo) poromechanics.

It can be solved by breaking the before_nonlinear_iteration routine into 4 different sub-routines, with each sub-routine having the respective method defined in the base sol. strategy.

The 4 sub-routines are:

1. Setting/ updating discretization parameters: already exists as SolutionStrategy.set_discretization_parameters
2. re-discretizing fluxes: A method SolutionStrategy.rediscretize_fluxes (or maybe more general SolutionStrategy.rediscretize_elliptic_discretizations?)
3. Copy results from first discretization to where it is required for the next step
4. re-discretize upwinding: already exists as SolutionStrategy.rediscretize, but should in this case be renamed to sth. like SolutionStrategy.rediscretize_hyperbolic_discretizations?

With these sub-routines, the base method SolutionStrategy.before_nonlinear_iteration would have the following unchanging form

class SolutionStrategy:

    def before_nonlinear_iteration(self) -> None:
        self.set_discretization_parameters()
        self.rediscretize_elliptic_discretizations()
        self.process_elliptic_discretization()  # Or any other name, but here we process the results of the new flux values
        self.rediscretize_hyperbolic_discretizations()

The models SinglePhaseFlow, EnergyBalance, MomentumBalance, Poromechanics, ThermoPoroMechanics would than
provide overrides of the sub-routines, and not of before_nonlinear_iteration.

The nice thing about this solution is, that the whatever happens in those sub-routines, its orthogonal in the sense it does not depend on what is happening in the other model (keyword: super-usage).

The changes would involve elevating some functionalities of compositional_flow.SolutionStrategyNonlinearMPFA to the level of SolutionStrategy, meaning large parts of the changes proposed here are already present in the core code (just at places where the order causes issues).

Now, by book-keeping the different non-linear discretizations in separate structures (for elliptic and hyperbolic), this change would not create an overhead at all. This can be achieved by introducing add_nonlinear_elliptic_discretization and add_nonlinear_hyperbolic_discretization instead of just add_nonlinear_discretization, meaning (flux) re-discretization is always optional and every model can define what needs to be re-discretized (like it is currently only done for Upwinding, while the poromechanics models sneak in some flux-rediscretization due to aperture change).
The change is solely a restructuring of routines.

@keileg let me know what you think.