Behaviour commands: terrain-relative tilt and command limits

Two more probes on the terrain loss of the default walk, each changing one thing. Neither brings back the 4 cm stairs.

• Probe 1: body tilt measured against the ground the feet stand on where it rises ahead, against gravity elsewhere; probe 2: the same, with commands on stairs and boxes limited to step height + swing ≤ 8 cm and ≤ 3 Hz
• Default walk, 100 trials, 4 cm stairs / 20° slope: previous probe 28 / 42, probe 1 32 / 71, probe 2 12 / 44, multi-terrain policy without behaviour training 86 / 100
• Terrain-relative tilt brings back part of the slope (42 → 71) and the 7 cm curb (40 → 58), not the 4 cm stairs
• Command limits lowered every hard test against probe 1 (4 cm stairs −20, slope −27), though their training curriculum climbed higher
• A separate difficulty level for the default walk and for commanded behaviours, tested the same day, did worse: 4 cm stairs 20 against 40