feat: human debug mode

[LufyCZ]

Description

This PR adds a simple way to quickly test text-based environments with human input, instead of having to wait for a hosted run to fail/reach the edge-case we want to test/...

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

Medium Risk
Touches the eval execution path and client selection, so regressions could affect how evaluations run (especially concurrency/TUI behavior), though the new mode is opt-in and well-tested.

Overview
Adds a new human debug mode for prime eval run (via --human-debug or human_debug=true in eval TOML) that replaces model API calls with terminal-entered assistant responses using a new HumanCLIClient.

When enabled, the CLI enforces a safer runtime (single-eval only, forces max_concurrent=1 and independent_scoring=true), disables TUI execution, and handles Ctrl-C cleanly; docs, an example config (configs/eval/human-debug.toml), and tests are updated to cover the new flag/client behavior and TOML validation.

^{Written by Cursor Bugbot for commit 6aa99ca. This will update automatically on new commits. Configure here.}