AI: Pinocchio auto-rigging (classical, native C++)

[fernandotonon]

Parent epic: #397

Goal

Predict skeleton placement for an unrigged humanoid/quadruped mesh using Pinocchio (Baran & Popović, 2007 — classical, no ML). Pairs with the libigl skinning weights task to give a complete unrig→rig→skin pipeline.

Scope

• Vendor Pinocchio's source (small C++ codebase, vendored to keep build deterministic).
• CLI: qtmesh rig model.fbx --skeleton humanoid -o rigged.fbx.
• Bundle skeleton templates: humanoid, quadruped, biped, generic.
• GUI: Animation Mode action "Auto-rig" with skeleton template picker and preview.
• MCP: auto_rig tool with template param.
• Output skeleton compatible with the existing animation pipeline.
• Pipe directly into AI: libigl bounded biharmonic skinning weights #402 (libigl skin weights) for a one-click "rig + skin" flow.

Acceptance Criteria

• Humanoid template produces a usable skeleton on a standard test character.
• Quadruped template works on a four-legged test asset.
• Output skeleton can be animated through the existing Animation Mode.
• Documented quality limitations (Pinocchio is 2007 — works best on roughly upright, manifold meshes).
• Sentry breadcrumb ai.assist.auto_rig recorded.

Effort

~7–10 days.