Fix uninitialized UVs, missing CUDA_CHECK in cleanup, face-skip warning, and unused alias

[Copilot]

Five unresolved review issues from the texture/MTL PR, all in the GPU scene build and OBJ loading paths.

Changes

• scene_description.hpp — addTriangle() and addTriangleWithNormals() set has_uvs = false but left uv0/uv1/uv2 uninitialized. Now explicitly zero-initialized to Vec3(0,0,0).

• scene_builder_cuda.cu — convertGeometry() — UVs were unconditionally copied from GeometryDesc even when has_uvs = false, reading uninitialized memory into GPU buffers. Now guarded:

  if (desc.data.triangle.has_uvs) {
      geom.data.triangle.uv0 = f2(...);
      // ...
  } else {
      geom.data.triangle.uv0 = f2(0.0f, 0.0f);
      // ...
  }

• scene_builder_cuda.cu — freeGPUScene() — Bare CUDA calls in the texture teardown path (cudaMemcpy, cudaGetTextureObjectResourceDesc, cudaDestroyTextureObject, cudaFreeArray, cudaFree) were not error-checked. All wrapped with CUDA_CHECK for consistency.

• obj_loader.hpp — Faces encountered before any usemtl (with no fallback) were silently skipped despite the docstring claiming a warning was emitted. Added a per-call face_skip_warned flag that emits one std::cerr diagnostic per loadOBJ() invocation, and updated the docstring to match.

• generate_uv_models.py — N = normal = (0, 1, 0) left normal unused (N was the actual use site). Simplified to normal = (0, 1, 0) with the one use site updated accordingly.

---

📱 Kick off Copilot coding agent tasks wherever you are with GitHub Mobile, available on iOS and Android.

[github-actions]

Docs out of sync?

This PR changes 3 source file(s) but no pages under website/docs/ were updated.

If this is a significant change (new feature, new CLI flag, new material/geometry type, performance improvement), please update the relevant documentation page. See the mapping in .github/copilot-instructions.md.

If this change genuinely doesn't affect the docs, you can ignore this message.