Missing model_configs/serve2.json and serving orchestration docs

[anselmotalotta]

• evaluation/run_hle.py expects model_configs/serve2.json and live retrieval/vLLM services on ports 1401–1407, but the repo lacks the config and instructions to generate it.
• Fix request: include model_configs/serve2.json template or a generator, and document how to start retrieval/vLLM services with the required models.

[shizhediao]

This file will be generated after starting the evaluation.