Add the test cases of the article, and the option that picks one

[gouarin]

Lot 2 of the plan: the monofluid cases of the article, and the machinery a case
needs once it comes in variants.

The cases

lax_liu is the four-quadrant Riemann problem, all nineteen Lax & Liu
configurations behind --riemann-config, interfaces at 0.8. In three
dimensions it is the eight-octant extension of configuration 3, the one the
article extends. It replaces riemann2d_config3, _config4 and _config12:
those differed by twenty numbers and nothing else, and every configuration
anyone wanted next was twenty more numbers to copy.

sod_x is the axis-aligned tube, the layout the article runs and the one
to hold against the exact solution with nothing projected in the way.
Registered in 1D, 2D and 3D. The rotated sod stays; it measures isotropy,
which the article does not.

blast_periodic is the initial condition of the weak scaling study: a high
pressure sphere at gamma = 5/3 in a periodic box. First case here to use
either.

triple_point_single_gamma is the geometry of section 6.1.2 with one gas
instead of two. Its header says at length that this is not the case of the
article, which needs the five-equation model. What earns it a place is the
longest run in the repository, on the most strongly adapted mesh.

Options that belong to a case

A case that comes in variants declares its own command line option. The
registry carries an options function, main() calls it before parsing, and
the case keeps the variable the parser writes into.

The option is declared by the binary it belongs to, so --riemann-config
accepts 1 to 19 in euler_2d and only 3 in euler_3d: a number the case
cannot honour is a parse error rather than a surprise at run time.

Every case's options are declared, not only those of the selected one, since
which case runs is itself decided by the parse. Two cases must therefore not
ask for the same option name. Naming the option after the case is the
convention that keeps them apart, and a collision is a CLI11 exception at
startup rather than a silent shadowing.

Three corrections to the reference table

The table here is Kurganov and Tadmor's, cross-checked against two independent
implementations. It differs from the article's reference code in three places,
noted in the header of lax_liu.hpp:

	reference code	Lax & Liu
config 6, upper left	p = 0.5	p = 1, as in all four quadrants
config 10, lower right	v = -0.4259	v = -0.4297 (the first is config 9's figure)
config 19	a copy of config 1	(1,0,0.3,1) / (2,0,-0.3,1) / (1.0625,0,0.2145,0.4) / (0.5197,0,-0.4259,0.4)

Neither of the two configurations the article uses is affected.

Renamed, not regenerated

The nine reference files of the three old Riemann cases are renamed. At
equal configuration the single case reproduces them bit for bit, which is what
says the regrouping changed no result. Checked before the rename, for all three
configurations and all three Riemann solvers.

Tests

The diagonal symmetry check now covers the six configurations invariant
under (x,y,u,v) -> (y,x,v,u) instead of three.

sod_x is held to the exact Riemann solution at both orders, in L1 and in the
position of each wave. The L1 errors fall by about eight between the orders,
against four for a smooth solution: the second order also puts the contact on
fewer cells.

A periodic box is held to conservation of mass, energy and momentum, the last
of which a wall does not conserve. It pins the gas as well, since the energy is
reconstructed at the gamma the case declares and agrees only if the solver ran
on the same one. Run at 1.4 it misses by 8%.

New reference files for the new cases, in every dimension they are registered
for. 106 fast tests and 22 slow ones pass.

Run to the end

Each case was taken to the final time the article gives, well past what the
short regression runs reach:

run	time	cells	levels	density
lax_liu #3, 2D, l in [4,9], t_f = 0.8	103 s	151 480	5	[0.138, 1.749]
lax_liu #3, 3D, l in [2,6], t_f = 0.8	49 s	167 350	4	[0.138, 2.040]
blast_periodic, l in [4,9], t_f = 0.2	28 s	95 692	5	[0.153, 9.563]
triple_point_single_gamma, l in [4,8], t_f = 2.0	198 s	138 234	5	[0.125, 3.314]

All at order 2, pressure positive throughout.

Not in this lot

The sparsity index measured on lax_liu #3 is 58% against the 12% or so the
article reports with a Löhner criterion. That is the question lot 3 asks, not a
defect here.