No test runs generate_airfoils with a delta_range over a wrapped slice, so the per-deflection .dat files are unchecked

[1-Bort-1]

#294 pins that generate_airfoils fits the contour it is handed, but only for the delta=0 contour: its test leaves delta_range unset and compares 1.dat against the direct LeastSquaresFit of the handed contour. Everything written per deflection — airfoils/{id}_d<delta>.dat, and the _cp/_cf node tables that share their geometry — has no test on its geometry at all.

The gap matters because the delta files inherit the base fit rather than recomputing it: generate_airfoil_aero takes kulfan_to_coordinates(base) and runs deform_section per delta (rotate about crease_frac, re-shrink_wrap at clearance=0, refit). On the SK100 the nine blown-up sections carried their deflected files with them, and 10_d10.dat is the case worth knowing: it looks normalised because a large chord rescaled it, while its lower surface is a constant min_clearance/chord. The existing coverage would not catch that — test/airfoil_aero/test_airfoil_aero.jl checks a flap's extrema(x) against [0, 1] within 0.01, which a rescaled blowup passes.

What is wanted:

• A generate_airfoils run with a delta_range spanning both signs, over a contour that has been shrink-wrapped first (the adapters' actual input shape — test/airfoil_aero/data/test_airfoil.dat wrapped, as Fit the contour generate_airfoils is handed, not a second wrap of it #294's test does, needs no new fixture), asserting every written _d<delta>.dat against the deflected geometry it should be: thickness and chordwise extent of the same order as the base section, and the delta=0 file unchanged from the existing check.
• Thickness as the measure rather than the bounding box, and the flap assertions in the deform_section testset moved onto it too, so a section rescaled by its chord cannot pass.
• Ideally one case that would have failed before Fit the contour generate_airfoils is handed, not a second wrap of it #294, i.e. driven from a wrapped contour rather than an analytic Kulfan section, so the test is a regression on the double wrap across the whole delta grid and not only at zero.

The unregularised fit underneath is #295 and not in scope here; this is coverage for the route as it now stands.

[1-Bort-1]

Coverage only, no src change: stacking on #294 and adding a generate_airfoils run over a shrink-wrapped test_airfoil.dat with a delta_range spanning both signs, asserting every written airfoils/1_d.dat on thickness and chordwise extent against the fit of the contour handed in — so the double wrap is caught across the whole delta grid, not only at zero. The deform_section flap assertions get the same thickness measure, which a chord-rescaled section cannot pass. Leaving the unregularised fit (#295) alone.

[1-Bort-1]

Folded into review feedback on #294 (#294 (comment)) rather than carried as its own issue.