Add core ambient_occlusion binding (#282); fix tutorial exact_geodesic call (#284)

[alecjacobson]

Resolves two issues from the triage.

#282 — igl.ambient_occlusion missing

ambient_occlusion was only bound in the embree submodule
(igl.embree.ambient_occlusion), so igl.ambient_occlusion(...) raised
AttributeError — and pip wheels built without embree had it nowhere.

libigl ships an embree-free overload in core
(igl::ambient_occlusion(V, F, P, N, num_samples, S), which builds an AABB and
shoots rays with libigl's own intersector). This PR binds it as the core
igl.ambient_occlusion, so it's available in every wheel. The
embree-accelerated variant stays at igl.embree.ambient_occlusion.

Matches the exact call the reporter tried: igl.ambient_occlusion(V, F, P, N, num_samples).
Adds test_ambient_occlusion (convex mesh → outward normals occlude ~nothing;
inward normals become substantially occluded; deterministic).

#284 — exact_geodesic returns an empty array

The binding was fine; the chapter-1 tutorial was calling the old signature:

d = igl.exact_geodesic(v, f, vs, vt)   # old 4-arg form

The current signature is (V, F, VS, FS, VT, FT), so vt bound to FS and
VT/FT stayed empty → output length #VT + #FT == 0 → empty array. Fixed the
notebook cell to name the arguments:

d = igl.exact_geodesic(v, f, VS=vs, VT=vt)

Testing

Built and ran the full suite locally: 92 passed.

🤖 Generated with Claude Code