Resolve a bug about incompatible versions of sklearn and xgboost.

dmlc/xgboost#11021 provides more info.
Since I don't want to use a from-source version of xgboost, Ill have to downgrade sklearn.

pyvista.def

@@ -15,7 +15,7 @@ From: nvcr.io/nvidia/jax:24.10-py3
 
     export PIP_TMP=/tmp/pip-cache
     mkdir -p $PIP_TMP
-    pip install --cache-dir $PIP_TMP xgboost pyvista pymc numpyro pandas numpy scipy polars openpyxl pyarrow imageio seaborn tqdm relplot xlsxwriter upsetplot fastexcel
+    pip install --cache-dir $PIP_TMP xgboost pyvista pymc numpyro pandas numpy scipy polars openpyxl pyarrow imageio seaborn tqdm relplot xlsxwriter upsetplot fastexcel "scikit-learn<1.6"
 
 %test
     # pytensor and matplotlib need work folders