Add before/after figures to #294 (doubly-wrapped contour fix)

[1-Bort-1]

#294 is the case that motivated figures in PR cards, and it is still open, so retrofit it as the first instance. The card argues in numbers — cond(A) 1.07e3 → 2.52e11, y reaching -13013 on 9 of 45 sections — that a second shrink-wrap collapses the Kulfan fit. A reader cannot check that by reading the three-line diff, and the failure is entirely visual: the blown-up sections look nothing like an airfoil.

Wanted, using the existing exports in ext/VortexStepMethodMakieExt.jl where they fit (plot_airfoil_fit, plot_airfoils, plot_slices_3d, plot_section_polars) and new functions there only if they are worth shipping to users:

• A 2D slice comparison for the SK100 sections that break — at least one of {6,8,9,10,12,16,27,29,43} — fit contour before the fix beside fit contour after, on identical axes, against the input contour that was handed in. The card notes 10_d10.dat looks normalised because it was rescaled by a large chord, so plot thickness rather than relying on the bounding box, or the figure will hide exactly the case the text warns about.
• Section polars for the same sections, before and after, across several delta, since the fix moves generated polars for every user of both adapters and that is the change's real risk.

The before is rendered from the merge base in a separate worktree with the same inputs and the same axes limits as the after. Publish through the assets mechanism from 1-Bart-1/Agents once it exists; if that is not ready, say so on the PR and hold rather than committing PNGs to the source branch. Update .agent/card.md so the figures land in the description with a one-line caption each, and leave the existing text intact — the figures are evidence for the argument, not a replacement for it.

[1-Bort-1]

Held rather than built: the assets mechanism (1-Bart-1/Agents#226) does not exist yet, so there is no URL a figure could use, and AGENTS.md §9 rules out committing PNGs. Said so on #294 and reported two further blockers there and here. Nothing pushed, no PR.

[1-Bort-1]

Held rather than built, and nothing was pushed. The issue's own fallback applies — the assets mechanism does not exist yet — but two further blockers turned up that outlast it, so this is the report rather than a one-liner. Said on #294 (comment 5632321550) and, for the part that belongs there, on 1-Bart-1/Agents#226 (comment 5632323139).

There is no host. 1-Bart-1/Agents#226 — orphan assets branch, publish helper, the AGENTS.md §6 rule — is open and still agent:needs-plan-approval, its session idle waiting on that. No branch named assets or figs exists on this repo (78 branches, paginated) or on Agents, and the agent CLI has no publish command. So a Markdown image has no URL to point at, and AGENTS.md §9 rules out committing the PNGs instead.

The subject is private and the host would be public. The SK100 mesh is not in this repo: it is data/sk100/obj/SK100SA_3d.obj in 1-Bart-1/BeyondTheSim.jl, which is private, while OpenSourceAWE/VortexStepMethod.jl is public. #226 pins every figure URL to a commit SHA and never rewrites it, and Camo caches hard enough that deleting the branch does not take it back — so rendering "the real case" publishes kite geometry to a permanent public URL. That is a decision to make before the first figure, not after, and it is not mine. If the answer is no, the honest move is a card that says so in words: data/ram_air_kite/ram_air_kite_body.obj is public and here, but the double wrap only collapses nine particular SK100 sections, so a stand-in would show a fix with nothing visibly wrong on either side.

Neither figure is producible with the exports as they stand. plot_airfoils(geometry_file; idxs=…) draws exactly the right content — the fitted line over the _raw.dat points — but has no limits keyword and autoscales each subplot under aspect=DataAspect(), so a section at 1e4 and a clean one land on different axes, which is the comparison #226 itself says is not evidence. plot_section_polars filters on POLAR_VECTORS and errors out on the POLAR_MATRICES grid that any delta_range run writes, plots every section with no selector, exposes no limits, and returns a PlotX rather than a figure or an axis. Nothing exported draws coefficients against α at several δ in 2-D at all: plot_polar_data is the only thing that reads the (α, δ) matrix and it is three 3-D wireframes of panels[1]. So both figures need new exported API — docstring, docs/src page, changelog — which is a PR of its own and a second decision, since the issue asks for new functions only where they are worth shipping to users.

Everything the next turn needs is in .agent/plan.md, including the two things that would otherwise be got wrong: the contour handed in is never written to disk and has to be recovered as shrink_wrap(x_raw, y_raw, ShrinkWrap(clearance=0.0, min_concave_radius=0.2)) — the SK100's wrap, not a bare ShrinkWrap() — and the .dat contents do not depend on α, so alpha_range=-2:2:2 reproduces them bit-identically at a twelfth of the cost while keeping the full delta_range. Generation is not the expensive part: the whole 45-section loop is about two seconds, since the solver is NeuralFoil batched over α, not Xfoil.

Opened #298 for something found on the way, and not worked around: generated_slices in ext/VortexStepMethodMakieExt.jl:1614-1615 builds its own filename tag for the per-deflection .dat instead of calling AirfoilAero.delta_suffix, the function that named the file when it was written. They agree only on positive integer degrees — measured: δ=-3 writes _dm3.dat and is looked for as _d-3.dat, δ=2.5 writes _d2p5.dat against _d2.5.dat, δ=-40 writes _dm40.dat against _d-40.dat. So plot_slices_3d silently draws no deflected contour for four of the SK100's eight non-zero deflections, and the warning it prints states the wrong scheme. The work following #294 generates both signs of δ, so it starts firing now.

Verification

• Blockers established by command, not inference: gh api --paginate .../branches on both repos, gh api repos/<r> --jq .visibility, agent --help, gh issue view 226 --repo 1-Bart-1/Agents
• plot_slices_3d builds its own deflection filename tag, so it finds no  #298's claim measured in the worktree's session, not read off the source: delta_suffix(deg2rad(δ)) against the ext's tag for δ ∈ {10, -3, 2.5, -0.5, 40, -40}, matching only at 10 and 40
• No test run and no code changed — this turn wrote nothing to the branch
• Risk: if Bart is content to publish SK100 section shapes publicly, blocker 2 evaporates and only Bug in untested branch causes SymbolicAWEModels.jl to fail #226 and the new plotting API stand between here and the figures.