Release v0.3.4

Features

• Introduced fully Differentiable Marching Tetrahedra (marching_tetrahedra.py) algorithm natively within conquer3d.ops.
• Added robust Delaunay background grid generation utilizing scipy.spatial with detached gradient tracking.
• Added 23 new native Stanford 3D repository asset classes (Bunny, Dragon, Armadillo, etc.) under conquer3d.data.assets that automatically stream geometry from reliable GitHub raw proxies.
• Added read_off and read_obj parsers in conquer3d.io with support for vertex RGB extraction.
• Optimized core CUDA routines for Marching Cubes (mc.cu, mc.h).

Refactors & Build

• Updated pyproject.toml and setup.py dependencies (including tetgen compatibility).
• Added test_data/ to .gitignore to prevent scratch downloads from polluting git history.
• Expanded MANIFEST.in to properly track new Python packages.

Examples

• Added diff_marching_tetrahedra.ipynb notebook implementing advanced TetWeave features: dynamic point optimization, SDF regularization, topological tracking, and interactive plotly wireframe renderings.
• Added diff_color_marching_tetrahedra.ipynb demonstrating simultaneous geometry and rainbow color-gradient optimization.
• Updated baseline Marching Cubes notebooks to mirror new visualization logic.