Bump libigl and align sparse marching_cubes with dense E2V behavior

[Copilot]

This updates the libigl pin to 72217e00bae6a7e041221e187aafbdc793f905d2 and brings the sparse-grid igl.marching_cubes binding in line with the dense overload by returning the edge-to-vertex map (E2V).

• Dependency update

  • Bump the vendored libigl revision in CMakeLists.txt to 72217e00bae6a7e041221e187aafbdc793f905d2.

• Binding behavior

  • Change the sparse-grid marching_cubes(S, GV, GI, isovalue) binding to call the libigl overload that returns E2V.
  • Make the Python return signature consistent across dense and sparse variants:
    • dense: (V, F, E2V)
    • sparse: now also (V, F, E2V)

• API/docs

  • Update the sparse overload docstring to document E2V.
  • Clarify the unpacking example for the returned edge map.

• Tests

  • Update the sparse marching-cubes coverage to unpack and validate E2V.
  • Handle empty maps safely when converting the returned dictionary into a NumPy array.

Example:

V, F, E2V = igl.marching_cubes(unique_S, unique_corners, J, 0.0)

E2V_triplets = np.array(
    [[k & 0xFFFFFFFF, k >> 32, v] for k, v in E2V.items()],
    dtype=np.int64,
) if E2V else np.empty((0, 3), dtype=np.int64)