vwxyzjn/cleanrl ? reverse-engineered prompt

Build me a small Python project for deep reinforcement learning that feels really clean and easy to read.

I want each algorithm to live in its own single file, so someone can open one script and understand the whole thing without digging through a big framework. Start with the basics like PPO and DQN, and make it easy to add a few more later. It should be able to train on common Gym style environments, log progress to TensorBoard, save runs with a seed so results are repeatable, and optionally track experiments with Weights and Biases. If it helps, include simple support for recording gameplay videos and running on the command line with a few flags like env name, seed, and total timesteps.

Please set it up so I can install it and run an example quickly, and include a clear README with simple get started instructions. If you need to check the latest docs for Gymnasium or similar, look them up online.