[new] agentic rl env datasets

[burtenshaw]

This is a draft of RL Environments on the hub. Please could you review the proposed yaml spec and snippets.

We're still waiting for PRs on Harbor and Moon to make this work.

---

Note

Low Risk
Documentation and an example manifest only; no application or API implementation changes in this diff.

Overview
Introduces Environment Datasets on the Hub: dataset repos can ship a root environment.yaml (spec_version: hf-env-0.1) so the Hub can show an Environment badge, framework chips, usage snippets, and expose declarations via /api/datasets/{repo_id}/environments.

Adds docs/hub/datasets-environments.md documenting required/optional manifest fields and three frameworks—Harbor (task bundles), Verifiers (row-based tasksets), and OpenEnv (Space or container)—with example YAML and CLI/Python snippets. The Datasets section of docs/hub/_toctree.yml now links Environment Datasets after Streaming Datasets.

Includes a sample environment.yaml at the repo root (Terminal-Bench–style Harbor/Verifiers env plus an OpenEnv coding env) as a concrete spec example.

^{Reviewed by Cursor Bugbot for commit 5e94c19. Bugbot is set up for automated code reviews on this repo. Configure here.}

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.