<doc> update LLM evaluation results and model details

docs: update README leaderboard (14 models, 2026-03-27)

fix: include the one-time search case for task 04_10

fix: add two additional ground truth paths for google scholar search task

docs: add repository citation metadata

Use per-run unique ID to prevent parallel run collisions

clean run.sh examples

Revise the font of google drive link.

refactor: extract hardcoded judge model to JUDGE_MODEL env var

<doc> add eval details

Update .env

Fix global summary scoring and optional thinking config.

Update README with custom model endpoint warning

Update .env

Revert script/run.sh to previous version.