Farama-Foundation/Minari ? reverse-engineered prompt

Build me a Python library for offline reinforcement learning datasets, something like an offline version of Gymnasium, where I can load existing datasets, inspect them, and create new ones from an environment while I collect episodes.

I want a simple command line tool too, so I can list remote datasets, download one, list what I have locally, and show details for a dataset by name. In Python, I should be able to do something like load a dataset, loop through episodes, and access observations, actions, rewards, terminations, truncations, and info for each episode.

It should also let me wrap a Gymnasium environment with a data collector, run an agent or random policy, and then save the collected experience as a reusable dataset. Please include a few clear examples and make the API easy to understand. If you need to look up current docs online, go ahead.