tools: add setup preflight checker

[Derekko-web]

Summary

• add eval/check_setup.py to catch common setup blockers before launching benchmark containers
• check repo layout, .env/API keys, selected harness Docker image, workspace data, task category paths, and prepare-script commands
• document the preflight command in the README

Validation

• OPENROUTER_API_KEY=dummy BRAVE_API_KEY=dummy python3 eval/check_setup.py --agent-backend openclaw --category all --model openrouter/test/model --skip-docker --skip-data
• OPENROUTER_API_KEY=dummy python3 eval/check_setup.py --agent-backend codex --category 01_Productivity_Flow --model openrouter/test/model --skip-docker --skip-data
• python3 eval/check_setup.py --help
• python3 -m py_compile eval/check_setup.py
• python3 -m compileall -q eval src
• git diff --check

[LennoxDai]

Thanks! Two small suggestions:

1. hf, yt-dlp, ffmpeg, and modelscope are only needed by script/prepare.sh, not for running the benchmark itself.

2. BRAVE_API_KEY is currently only checked for all / 04_Search_Retrieval, but other categories may also rely on web search. Suggest checking it unconditionally.

Happy to send a follow-up PR if helpful.