1.0.14

• Add a --split flag to both CLIs (-sp/--split in C++, --split in Python): additionally writes each convex part to its own file in an <output>_parts folder. Needed by consumers that convexify each mesh asset as a whole (MuJoCo, Bullet, ...), where a single merged file always collapses to one convex shape (#53). The merged output is unchanged.
• Make decomposition and merging deterministic with respect to thread scheduling: merge-stage sampling is now seeded per candidate pair, and results are collected in a fixed order instead of thread-completion order (#43). Note: this changes default output slightly — rerunning with the same input and seed now reproduces the same decomposition regardless of thread count.