Add control-surface support to VortexLatticeMethod

[gaoflow]

Overview

This adds control-surface support to VortexLatticeMethod, which was previously unported to the differentiable VLM (#134). Control surfaces defined on the Airplane geometry — exactly as they already are for the AVL and LiftingLine backends — are now honored by the VLM.

wing = asb.Wing(
    name="Wing",
    symmetric=True,
    xsecs=[
        asb.WingXSec(
            xyz_le=[0, 0, 0], chord=1, airfoil=asb.Airfoil("naca0008"),
            control_surfaces=[
                asb.ControlSurface(name="flap", symmetric=True, hinge_point=0.75, deflection=10),
            ],
        ),
        asb.WingXSec(xyz_le=[0, 5, 0], chord=1, airfoil=asb.Airfoil("naca0008")),
    ],
)
aero = asb.VortexLatticeMethod(airplane=asb.Airplane(wings=[wing]), op_point=asb.OperatingPoint(velocity=10, alpha=0)).run()
# Before: deflection ignored (CL ~ 0).   After: CL > 0, as expected for a deflected flap.

Approach

In a thin-surface vortex-lattice method, a control-surface deflection enters the problem only through the flow-tangency boundary condition — i.e. by rotating the affected panels' normal vectors about their local (spanwise) hinge axis. The vortex lattice itself is held fixed. This is the standard linearized treatment of control surfaces in a VLM, and it keeps the implementation fully vectorized and differentiable.

Concretely:

• Panel → control-surface mapping. Reusing the same per-spanwise-strip bookkeeping that LiftingLine already uses (a control surface is associated with the inboard cross-section of each strip; symmetric surfaces deflect both sides of a symmetric wing identically, antisymmetric surfaces deflect the mirrored side oppositely). A new helper, _compute_control_surface_deflections(wing), returns a per-panel deflection angle in the exact panel order that Wing.mesh_thin_surface produces. A trailing-edge surface deflects every panel whose chordwise center lies aft of its hinge_point; a leading-edge surface (trailing_edge=False) deflects every panel forward of it.
• Applying the deflection. Each affected panel's normal is rotated toward the panel's aft chordwise direction by its deflection angle (n' = cos(δ)·n + sin(δ)·ĉ, with δ > 0 trailing-edge-down and ĉ the aft chordwise direction projected perpendicular to n, so the rotated normal stays a unit vector).
• The rotation branch is guarded by a purely structural check (does any control surface exist at all?), so airplanes with no control surfaces produce bit-for-bit identical results, and deflection magnitudes may be Opti variables for differentiable trim/control optimization.

Verification

Added four tests in test_vortex_lattice_method.py, all passing (full module: 8 passed):

• test_control_surface_zero_deflection_matches_baseline — a control surface at deflection=0 reproduces the no-control-surface result to within 1e-12 across CL, CD, Cl, Cm, Cn.
• test_control_surface_symmetric_flap_increases_lift — a downward symmetric flap monotonically increases CL (0 → 0.23 → 0.47 → 0.96 for δ = 0°, 5°, 10°, 20° at α=0) with zero net roll.
• test_control_surface_aileron_produces_roll — an antisymmetric aileron produces a nonzero rolling moment with zero net lift.
• test_control_surface_deflection_is_differentiable — the deflection angle solves as an Opti variable (trimming a flap to a target CL), confirming the analysis stays differentiable.

ruff check / ruff format --check clean on both changed files. The four pre-existing VLM tests still pass unchanged.

Note: I don't have an avl binary in this environment, so I verified against physical expectations (sign, monotonicity, symmetry, and exact baseline reproduction) rather than a numerical AVL cross-check. Happy to add an AVL comparison if useful.

Resolves #134.