staghuntrpg/agar ? reverse-engineered prompt

Build me a simple Agar.io style game environment in Python that I can use for reinforcement learning.

I want the core gameplay to match Agar.io, where each player controls one or more cells in a petri dish, moves toward the mouse position, eats smaller cells to grow, and can split into multiple cells when large enough. It should support multiple controlled agents at once, not just one, and it should be set up as an RL environment with reset and step style interaction. Please keep it focused on the free for all mode only.

Also add a basic human control mode so I can run it locally and play with mouse and keyboard to test that it works. If you need to check any current library details, look up the latest docs online. Keep the default settings sensible, and make sure the project runs cleanly with a simple command.