This removes the "Bogus distance" warnings that are otherwise generated
by `addZeroVelocityBoundary`.

i.e. output cell-local quality information by default

I am not quite happy with how this looks right now but at least both
validation examples are for the most part condensed to only differ in
their refinement setup.

Perhaps a SchaeferTurek-specific "Solver" would further improve
reproducibility?

i.e. cylinder2d_optimized_grid contains roughly as many grid cells as a
uniformly resolved N=12 lattice but models the cylinder in more detail
and is more stable.

Reproducing past results for new plots is otherwise getting quite annoying.

Weirdly it seems easier to get reasonably close values for the unsteady
flow. Results for the steady flow seem to get worse when increasing
grid resolution even in uniform grids.

Finally seems to stop mixing up material numbers for every other resolution I try it out with.
Cylinder diameter is now actually set to 0.1m as called for by [SchaeferTurek96].

Makes it easier to exclude refined areas

Required to both increase cylinder discretization detail and fix cell count to a uniform N=9 grid.

This way statistics reflect the cells disabled by refinement.

Mainly to further declutter simulation and refinement setup

Somewhat fiddled together but works as expected. Refinement areas are
informed by Knudsen metric results. The key is to refine the outflow.
Sadly this requires refinement-overlap-crossing boundary conditions.

Interface for detailed (i.e. cell-relative) grid refinement control
still has room for improvement. Mixing physical indicators and cell
alignment requirements is problematic.

This allows for readable differentiation between constructor overloads.

By deriving RefiningGrid2D from Grid2D to provide methods specific to refining subgrids.

Scaling factor is specific to the refinement method implemented by the coupler.

Works but is nowhere near anything one could consider good.
Obvious issues:

* More than one cuboid per grid makes it harder to determine the next lattice cell to be coupled
	* i.e. currently lattice positions are determined ad hoc by resolving their physical position
* Coupling is not actually parallelized
	* All coupling lines are traversed by all processes, way to much communication
* Load balancing and cuboid decomposition doesn't care about refinement
	* ideally refined cuboids should be computationally near their coarse _parent_ cuboids

The first two isses should be fixable with a reasonable amount of work.
This sadly doesn't apply in any form to the last issue.

Grid refinement is at its easiest when discrete materials are set using
analytical - i.e. resolution independent - indicators.

Coupler construction is to be wrapped by `Grid::refine` for refinement
areas that are fully inside the coarse domain.