Summary

• add --preflight-only to validate the resolved LLM endpoint/model/key before loading benchmark data
• allow preflight-only runs without MS MARCO paths or SQLite DB paths
• document the DeepSeek preflight step before the longer agent-loop run

Verification

• uv run pytest tests/test_agent_search_benchmarks.py tests/test_deepseek_env_config.py
• uv run ruff check examples/ablation/run_agent_loop_benchmarks.py examples/ablation/configure_deepseek_env.py tests/test_agent_search_benchmarks.py tests/test_deepseek_env_config.py
• uv run ruff format --check examples/ablation/run_agent_loop_benchmarks.py examples/ablation/configure_deepseek_env.py tests/test_agent_search_benchmarks.py tests/test_deepseek_env_config.py
• git diff --check
• uv run python examples/ablation/run_agent_loop_benchmarks.py --llm-preset deepseek --preflight-only (expected: DEEPSEEK_API_KEY is not set)