openai/gym ? reverse-engineered prompt

Build me a Python toolkit for reinforcement learning experiments, similar to Gym. I want researchers and students to be able to plug their learning code into different simulated environments without rewriting everything each time.

The main thing should be a simple environment interface where a user can create an environment, reset it, take actions, get back an observation, reward, whether the episode ended, and extra info. Include a small set of example environments so someone can run a random agent right away and see the loop working. Make environment names versioned so results can stay reproducible when an environment changes.

Please include clean install instructions, optional dependency groups for heavier environment families, basic tests, and a short README with a copy paste example. Keep it friendly for people new to reinforcement learning, but solid enough that experienced users can build their own environments on top of it. Look up current docs online if you need to.