Add 3D mesh watermarking toolkit

[gitlujean]

Motivation

• Provide a pure-Python toolkit to embed and extract a vertex-index-based digital watermark for 3D triangle/polygon meshes that is robust to common geometry-preserving attacks.
• Use a stable axial partitioning and index-relabelling approach so the embedding is geometry‑lossless (no coordinate changes) while remaining tolerant to similarity transforms, mild smoothing, noise and simple simplification.

Description

• Add a watermark3d package implementing the algorithm and utilities, including watermark.py (stable-axis PCA, rank-based interval assignment, embed_watermark, extract_watermark, bit_error_rate) and configuration via WatermarkConfig with a repetition code (repeats).
• Add mesh.py to load/save OBJ and ASCII PLY meshes while preserving OBJ face token suffixes and synchronously rewriting face indices after relabeling vertices.
• Add simple attack simulators in attacks.py (similarity transform, controlled noise, Laplacian smoothing, order-preserving vertex decimation) and a CLI watermark3d/cli.py exposing embed/extract operations with metadata and BER reporting.
• Add scripts/test_directory.py to run directory-level experiments (auto-generate a sample mesh if none provided) and produce watermark_report.json, and update README.md with algorithm details, usage and robustness notes.

Testing

• Compiled all modules with python -m py_compile watermark3d/*.py scripts/test_directory.py and the compilation succeeded.
• Ran the directory test runner on generated sample meshes with python scripts/test_directory.py --input-dir /tmp/wm_ci_in --output-dir /tmp/wm_ci_out --bits 1011001110001011 --repeats 3 --make-sample-if-empty and the produced report showed successful embedding and extraction (BER=0.0) for the final tuned settings.
• Verified single-file CLI flows with python -m watermark3d.cli embed ... and python -m watermark3d.cli extract ... which returned the expected bitstring and BER=0.0 in the tested cases.

---

Codex Task