Training on a lab PC with MuJoCo Warp

Training moved to a lab PC with an RTX 5070 Ti, and its physics from the JAX backend of MuJoCo MJX to the Warp backend. Rewards, randomization and the PPO setup are unchanged.

• Physics steps per second on flat ground: 0.90 M on JAX, 8.1 M on Warp
• On the stair field: 0.52 M on JAX, 4.09 M on Warp
• 1 billion training steps of the official-size policy: 46 min
• Training curves and the stair-course result match between the two backends