feat: validation fallback when an intermediate XFOIL run fails

[PSamuele]

What

Before, if one of the intermediate XFOIL runs failed, the validation showed only the total ("breakdown N/A").
Now each line uses the best breakdown allowed by the runs that worked. The parts always add up exactly to the total:

Failed run	Breakdown	Mark on the line
none	average of the two orders	none
viscous at Mach 0	Mach first: Mach = B - C, visc = C - D	(Mach first)
inviscid at the real Mach	viscosity first: visc = B - E, Mach = E - D	(visc first)
both of the above	Mach + visc = B - D	(Mach+visc combined)
inviscid at Mach 0	only the total	(breakdown N/A)

With a single order, the split between Mach and viscosity can differ from the average by a few thousandths of Cl (up to 0.006 measured), hence the mark.

Other changes:

• CSV: new columns Breakdown (method used) and dCl_Mach_plus_Viscosity.
• Error plot: single-order points have hollow markers; lines are interrupted where a value is missing instead of joining the points around the gap; fixed colours (the extra markers could shift the colour cycle between subplots).
• Two lines in the terminal header explain the marks. README updated.

Verified

• Windows: NACA 2412 at 6 deg, where XFOIL viscous at Mach 0 fails (in two runs out of two), now shows (Mach first): -0.0097 - 0.0155 + 0.0748 = +0.0496 = total. The other 23 lines are identical to the previous PR.
• Sandbox: failures simulated in 6 runs covering every level; the sum is exact in all cases with a breakdown. Without failures, all 24 lines are identical to the previous PR.
• Error plot checked visually.

Depends on

#11 (base branch: perf/vectorized-panel-method).