[v0.1]Docs: Model performance evaluation guide

[rootfs]

Acceptance

• Documents automated workflow to evaluate models (including but not limited to MMLU-Pro), generate performance-based routing config, and update categories[].model_scores
• Includes example evaluation->config pipeline.

[JaredforReal]

I would like to contribute to this issue. I have read code in /src/training/model_eval, and plan to complete this with two PRs:

1. arc_challenge_vllm_eval.py lacks enough comments to read, I will fix it by comparing it to model_eval/mmlu_pro_vllm_eval.py
2. Write a doc with example in each practical step:
   1. Intro: why model perf eval
   2. Prerequisites
   3. Evaluate on MMLU-Pro: mmlu_pro_vllm_eval.py
   4. Evaluate on ARC Challenge: arc_challenge_vllm_eval.py
   5. Visualize per-category performance: plot_category_accuracies.py
   6. Generate performance-based routing config: result_to_config.py and config/config.yaml

[Xunzhuo]

Cool @JaredforReal !

[JaredforReal]

@Xunzhuo Thanks for your time! I have made the first PR, and the Second one may take more than one day. I will work on it ASAP.
Looking forward to any suggestions!