Update to 2026-09-17 pipeline snapshot; add weights, L20 assets and recording via Git LFS

Source: RGB-D -> Dyn-HaMR -> L20 retargeting -> FoundationPose -> reference repair -> SPIDER,
documented in docs/PIPELINE_LATEST.md and docs/SETUP_AND_WEIGHTS.md. Adds FoundationPose and
nvdiffrast upstream snapshots, requirements/pipeline_venv.txt and the FoundationPose weight
manifest/downloader.

Assets (Git LFS): weights/ (WiLoR detector, HandFlow denoiser, UniDepth-L), FoundationPose
checkpoints, HaMeR checkpoint, Dyn-HaMR HMP model and BMC constraints, L20 URDF/meshes, the
20260915_171525 D405 recording and the two box CADs. MANO models are not redistributed
(third_party/hamer/_DATA/data/mano/README.txt). Environments, caches and run outputs excluded.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

scripts/red_box_distance.py

@@ -0,0 +1,18 @@
+"""Unsigned closest distance with solid-angle winding sign for a closed mesh.
+Avoid ray-parity misclassification at collinear CAD tessellation edges.
+"""
+import numpy as np
+import open3d as o3d
+
+def signed_distance(scene,mesh,points):
+ p=np.asarray(points,dtype=np.float64)
+ d=scene.compute_distance(o3d.core.Tensor(p.astype('float32'))).numpy()
+ ids=np.flatnonzero(((p>=mesh.bounds[0])&(p<=mesh.bounds[1])).all(1))
+ triangles=np.asarray(mesh.triangles,dtype=np.float64)
+ for start in range(0,len(ids),32):
+  ix=ids[start:start+32];v=triangles[None]-p[ix,None,None,:];a,b,c=v[:,:,0],v[:,:,1],v[:,:,2];length=np.linalg.norm(v,axis=-1)
+  numerator=np.einsum('bij,bij->bi',a,np.cross(b,c))
+  denominator=length.prod(-1)+(a*b).sum(-1)*length[:,:,2]+(b*c).sum(-1)*length[:,:,0]+(c*a).sum(-1)*length[:,:,1]
+  winding=np.arctan2(numerator,denominator).sum(-1)/(2*np.pi)
+  d[ix[np.abs(winding)>.5]]*=-1
+ return d