Add Terminal-Bench sandbox example environment

[willccbb]

Summary

• add a configurable SandboxEnv wrapper for Terminal-Bench tasks that stages assets in a fresh sandbox and runs the official tests during post-rollout
• cache the test outcome for scoring via a simple rubric and ship a single-example dataset for the requested task id
• document the environment and declare its dependencies

Testing

• uv run ruff check environments/terminal_bench/terminal_bench.py

---

https://chatgpt.com/codex/tasks/task_e_68ea1a482a388326932b1fb2fa1d666d