Farama-Foundation/Gymnasium ? reverse-engineered prompt

Build me a Python library for reinforcement learning that gives people a simple, standard way to talk to environments.

I want it to feel easy to use, so someone can install it, make an environment like CartPole, reset it with a seed, take random actions, and get back observations, rewards, and whether the episode ended. Include a few built in sample environments that cover simple toy problems, classic control, and a couple physics based ones if possible.

Please also add clear docs and a few basic tests, so it is easy to understand how to use and hard to break. Make the API clean and consistent, with environment versioning so changes don’t mess up results. If you need to check current library details or examples online, go ahead and look them up.