[nicolas-maman]

Toward muscles as the default (#790): two switches to run everything on them, and level feet for figures with ankles.

Switches (read once, as a figure or ragdoll is made):

• AE3D_MUSCLES=1 starts every active ragdoll on muscles (motion.set_muscles).
• AE3D_ANKLES=1 gives every ragdoll feet of its own (physics_set_ankles).

With both, any suite, the showcase or a game runs on muscles unchanged, which is how the suites still failing there are found.

Level feet for ankled figures (#777). Each foot was placed at its own toe's height along its shin. The box man's right toe sits 8 mm under its left, so it stood its whole weight on the right foot (844 N against 0): every step began from one loaded foot, and on muscles a figure tripped onto its front never got up. Both feet now sit at the lower toe's share of the shin:

• test_get_up on muscles and ankles: 36 of 36 (3 failures before).
• The figure keeps its height, and so its weight: ragdoll_set_mass weighs a figure by its height down to its lowest foot. Levelled at the mean of the two toes instead, the box man came out 4 mm shorter and 0.36 kg lighter than without ankles, which test_ankled_steps caught.
• A figure without ankles keeps each foot at its own toe's until its protective fall is retuned.

A knockdown falls back, not over (#822). Siltreach's knock (2.9 m/s back and 0.25 up at the chest) sent its muscled figure over its shoulders with its legs over its head, and it rose turned half round. The protective fall was swept with the balance assist held against the fall until the figure is down (FALL_ASSIST), standing in for the legs and hips a person brakes a fall with. On muscles the knock's unpin let go of that brake too. A fall now keeps it until the figure lands (pin_held); the pin is still let go standing, lying and through the get-up. On Siltreach's own figure in its idle:

• the hips rise 0.09 m over the neck at the most, against 0.42;
• it rises turned 0.27 rad, against 2.71;
• it stays clean up to 3.0 m/s (the servo up to 3.2).

A crash fixed. physics.ragdoll_balance_assist asked the parallel joint for its torque while the pelvis anchor was the motor joint that carries a figure following its clip (animated, getting up), and the program crashed. It answers 0 there now (aephysics#181 has the getter side).

docs/motion.md: a section on muscles: what a figure on them gets, how to switch them on, and what still fails there.

Tested locally (Windows, Aether 0.831.0): test_ankled_steps 3/3, test_muscles 11/11 (new: knocked down backwards, the brake holds and the hips stay under the neck; both fail without the fix), test_ragdoll_mass 9/9 (new: an animated figure's assist reads 0; a segfault without the fix), test_get_up 36/36 (and 36/36 with AE3D_MUSCLES=1 AE3D_ANKLES=1), test_get_up_kerb 24/24, test_feet 20/20.

Closes #822. Part of #790 and #777.