update call to `mean_squared_error` to comply with sklearn v1.6

The `squared` argument of the scimitar-learn method `mean_squared_error` is depreciated since v1.4 and has been removed in v1.6, making this evaluation module incompatible with the latest version.
This PR makes it call the `root_mean_squared_error` when `squared=True`, solving this error.

For reference: https://scikit-learn.org/1.5/modules/generated/sklearn.metrics.mean_squared_error.html (docs of v1.5)

cc

@lvwerra

mse.py

@@ -14,7 +14,7 @@
 """MSE - Mean Squared Error Metric"""
 
 import datasets
-from sklearn.metrics import mean_squared_error
+from sklearn.metrics import mean_squared_error, root_mean_squared_error
 
 import evaluate
 
@@ -112,8 +112,13 @@ class Mse(evaluate.Metric):
 
     def _compute(self, predictions, references, sample_weight=None, multioutput="uniform_average", squared=True):
 
-        mse = mean_squared_error(
-            references, predictions, sample_weight=sample_weight, multioutput=multioutput, squared=squared
-        )
+        if squared:
+            mse = mean_squared_error(
+                references, predictions, sample_weight=sample_weight, multioutput=multioutput
+            )
+        else:
+            mse = root_mean_squared_error(
+                references, predictions, sample_weight=sample_weight, multioutput=multioutput
+            )
 
         return {"mse": mse}