fix: restore UI, add /tasks /grade, clamp scores to (0.05, 0.95)

Massive Rewrite: Rebuilt Phase 2 standalone validation constraints perfectly mapped to 0.05-0.95

Rewrite environment, app endpoints, and openenv.yaml for Phase 2 validation

Fix phase 2 check: update default task_score to 0.0001 for observation fallbacks

hf

good to go

feat: implement SQL agent environment with OpenEnv integration, SQLite management, and automated task grading