[DON'T MERGE] Update the eval procedure & Test the Evaluation Workflow

[yl231]

Summary

Updates the evaluation pipeline for public dataset support and adds multi-threaded evaluation.

Key changes:

• Multi-threaded evaluation with configurable workers (--num-workers)
• Migrated to public RouteWorks/RouterArena dataset (removed HF_TOKEN)
• Fixed column name handling for dataset compatibility
• Added glm-4-air-router submission

Pre-commit checks pass. Ready for evaluation workflow.

[github-actions]

📊 Router Evaluation Results

Router: glm-4-air-router
Dataset Split: full

Metrics

Metric	Value
RouterArena Score	0.5617
Accuracy	54.65%
Total Cost	$0.396728
Avg Cost per Query	$0.000047
Avg Cost per 1K Queries	$0.0472
Number of Queries	8400

---

Evaluation completed by RouterArena automated workflow

[github-actions]

📊 Router Evaluation Results

Router: glm-4-air-router
Dataset Split: full

Metrics

Metric	Value
RouterArena Score	0.5617
Accuracy	54.65%
Total Cost	$0.396728
Avg Cost per Query	$0.000047
Avg Cost per 1K Queries	$0.0472
Number of Queries	8400

---

Evaluation completed by RouterArena automated workflow

[yl231]

Test passed so I will close this test PR.

[github-actions]

📊 Router Evaluation Results

Router: glm-4-air-router
Dataset Split: full

Metrics

Metric	Value
RouterArena Score	0.5617
Accuracy	54.65%
Total Cost	$0.396728
Avg Cost per Query	$0.000047
Avg Cost per 1K Queries	$0.0472
Number of Queries	8400

---

Evaluation completed by RouterArena automated workflow