Feature/voronoi safe fallback

[Jammy2211]

This pull request improves the robustness of the Voronoi area calculation in the scipy_delaunay function by adding error handling. Now, if the Voronoi computation fails (for example, due to Qhull precision errors), the code will gracefully fall back to using barycentric dual areas instead of crashing.

This is because the new border relocation scheme could sometimes create source pixel distributions which broke the Voronoi code, albeit this happened rarely for odd mass models.

Error handling and fallback logic:

• Added a try/except block around the call to voronoi_areas_numpy(points) in scipy_delaunay, so that if an exception occurs (such as a Qhull precision problem), the code prints an informative message and falls back to calculating areas using barycentric_dual_area_from(points, simplices, xp=np) instead.