Improve vf-eval progress display for multi-env metrics

[snimu]

Description

Show metrics during vf-eval rollout. On a narrow screen:

On a wide screen (before the first results come in):

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

Note

Introduces a richer progress display for vf-eval with multi-environment awareness and live metric tracking.

• Adds verifiers/utils/progress_utils.py with MultiEnvProgress and RolloutProgress to render per-env progress and rolling averages of reward and rubric metrics; includes TTY detection via can_render_rich_progress() and falls back to tqdm.
• Updates Environment.generate to precompute per-task totals, discover metric names via env_map or fallback, initialize rich progress (multi-env or single), call update(states) as tasks complete, and cleanly start/stop; removes manual reward averaging/postfix logic.
• Minor typing/cleanup adjustments in data_utils.py and message_utils.py (casts, removed type ignores).

^{Written by Cursor Bugbot for commit 6d3a02a. This will update automatically on new commits. Configure here.}

[cursor]

Tqdm fallback no longer updates rolling average reward

Medium Severity

When the rich progress display cannot render (e.g., non-TTY environments like CI/CD), the code falls back to tqdm. The tqdm progress bar is initialized with postfix=dict(reward="?"), but the loop only calls pbar.update(1) without ever calling pbar.set_postfix() to update the reward display. The old code tracked reward_sum and reward_count and updated the postfix with the rolling average reward. Now users in non-TTY environments see "reward=?" throughout the entire evaluation, which is misleading and a regression from the previous behavior.

Additional Locations (1)

• verifiers/envs/environment.py#L961-L962