motion.hold: the drawn hands are on the grips but the forearm bodies are 13 to 50 cm off them (held off by the thighs) #843

Siltreach measured its seven settlers eating on a 0.38 m log with motion.hold (#839) on ae3d e399f6a. The grips are at (±0.09, 0.08, 0) on a 0.11 m bowl, with a bite every 4 s.

What's wrong: the drawn hands are on the grips, but the forearm bodies aren't.

• The drawn wrist (hand_l/hand_r) is 0.4 to 1.0 cm from its grip.
• Each forearm body's far end (ragdoll_body_pose then ragdoll_pose_end(BONE_LOWER_ARM_L/R)) is 13 to 16 cm away, on every sitter. For example, Kade is at L 0.130 and R 0.158.
• On two sitters the right forearm body is much further off and gets worse through the meal: Wren reaches 0.36 to 0.38 m by 4.6 s, and Brannoc sits at 0.50 m from 9 s to 57 s. Their left forearms stay at 1.6 to 4.4 cm.
• The bowl follows its own plan: it is 11.18 cm from mouth() on all 98 bites.

Likely cause: with the bowl over the lap, the plan has the forearms passing through the thighs. Forearms collide with their own thighs and pelvis; aephysics' no-collide group is only spine_01 and the thighs. So the bodies are held off where the drawn arms go. This is the same family as #700 (reach around the trunk): the planned limbs must clear the figure's own bodies.

Why it matters: while acting, the figure is drawn from the plan, so it looks right. But the bodies are what gets hit and what takes over at the hand-over, and Siltreach holds its sitters' bodies to 3 cm.

Wanted: a hold whose forearm bodies reach their grips (under 3 cm) for the whole hold, bites included. tests/test_eat.ae should also check the bodies, not only the drawn palm.

Repro: sit plus hold on a 0.38 m log, the grips as above, a 60 s hold with a bite every 4 s, measuring both forearms' pose_end from the body pose against the grips.