feat(bench): HMMT/AIME small-subset harness + answer extraction tests

conftest.py

@@ -0,0 +1,7 @@
+"""Root conftest : add repo root to sys.path so `from scripts.* import ...` works."""
+import sys
+from pathlib import Path
+
+_ROOT = Path(__file__).resolve().parent
+if str(_ROOT) not in sys.path:
+    sys.path.insert(0, str(_ROOT))