[evals] Update pyproject.toml, forcing datasets<4.0.0 (#118)

`datasets` package is officially updated to 4.0.0 on 2025.07.10. 
scripts are no longer supported in 4.0.0, but this is used in some benchmarks (e.g., LiveCodeBench)
So `datasets` version is forced "<4.0.0".

Author: zhouxiangxin1998

pyproject.toml

@@ -10,7 +10,7 @@ dependencies = [
     "vllm==0.7.0",
     "word2number",
     "scipy",
-    "datasets",
+    "datasets<4.0.0",
     "latex2sympy2",
     "pydantic",
     "setuptools",