track token usage in eval

[willccbb]

Description

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

Additional Notes

---

Note

Medium Risk
Touches core eval progress/saving paths and changes concurrency behavior; mistakes could skew reported usage/metrics or alter throughput, but the changes are additive and largely display/metadata-focused.

Overview
Evaluation runs now capture token usage by summing response.usage across trajectory steps into per-rollout token_usage, and computing average usage (input/output tokens) in GenerateMetadata when saving results.

Live and post-run displays are updated to show token usage: the Rich eval display tracks rolling average tokens and conditionally adds input/output columns to the final summary, and the results-view TUI shows avg input/output tokens in the metadata panel. Numeric formatting is centralized via a new format_numeric helper, and concurrency handling is tweaked to scale max_concurrent when group scoring is used (non-independent_scoring with multiple rollouts).

^{Written by Cursor Bugbot for commit 6446b78. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[samsja]

lgtm