Fix ORTEP arc boundary: use ellipsoid silhouette normal condition instead of Pz=0

[Copilot]

Root cause

In _draw_principal_arcs the boundary between the solid (front) and absent (back) halves of each principal great-circle arc was determined by where the z-coordinate of the 3D point on the arc is zero (Pz = 0).

The correct ORTEP criterion is where the outward normal of the ellipsoid is perpendicular to the viewing direction, i.e. (U⁻¹P)_z = 0.

For a point P(t) = rᵢ·vᵢ·cos(t) + rⱼ·vⱼ·sin(t) lying on the ellipsoid surface:

(U⁻¹P)_z = (vᵢ[2] / rᵢ) · cos(t)  +  (vⱼ[2] / rⱼ) · sin(t)

Setting this to zero gives phi_n = atan2(vⱼ[2]·rᵢ, vᵢ[2]·rⱼ).

The old code computed phi_z = atan2(rⱼ·vⱼ[2], rᵢ·vᵢ[2]) — the roles of rᵢ and rⱼ are swapped compared to the correct formula.

Effect

Condition	Arc endpoints
Old (Pz = 0)	Fall inside the 2-D projected ellipse for anisotropic atoms
New (silhouette normal)	Land exactly on the projected ellipse boundary ✓

The two conditions are identical for isotropic ellipsoids (rᵢ = rⱼ), which is why only some ellipsoids showed the wrong curves.

Fix

A 2-line change in _draw_principal_arcs:

# Before
Az = ri_3d * vi2
Bz = rj_3d * vj2

# After – swap ri_3d ↔ rj_3d so the atan2 encodes (U⁻¹P)_z = 0
Az_n = vi2 * rj_3d
Bz_n = vj2 * ri_3d

Tests

192 tests pass (1 pre-existing failure due to missing PIL dependency, unrelated to this change).