Reproducing experiments when using the environment

Use uv sync --frozen to install the base versions recorded in uv.lock. For PPO training and evaluation, install the example and RL extras as well:

uv sync --frozen --extra examples --extra rl

The examples create all simulation randomness from explicit JAX PRNG keys, so reruns are deterministic when the same seed, environment ID, and dependency versions are used.

The PPO examples save and load model parameters from trained_models_ppo/. For evaluation, keep the TrainConfig.env_name and TrainConfig.run_name values aligned with the model file names in that directory.