TorchRL-PPO

Build reinforcement-learning training code with TorchRL APIs, especially PPO agents and examples using Gym/Gymnasium environments. Use when implementing or reviewing TorchRL code that should rely on TensorDict, GymEnv, TransformedEnv, TensorDictModule, ProbabilisticActor, ValueOperator, SyncDataCollector, ReplayBuffer, GAE, and ClipPPOLoss instead of hand-writing PPO losses, rollout storage, or actor-critic plumbing.