fix(training): provide more informative error when user specifies inexistent node attribute (#663)

Small fix to make the error message more informative. Instead of just
telling a user that the specified attribute does not exist in the graph
data, it also provides a list of the ones that are available. So for
instance, if one mistakenly specified "cutout_mask" instead of "cutout",
they will know right away what the problem is.

Author: frazane

training/src/anemoi/training/losses/scalers/node_attributes.py

@@ -94,8 +94,11 @@ def __init__(
             norm=norm,
             **kwargs,
         )
-        if self.scaling_mask_attribute_name not in self.nodes:
-            error_msg = f"scaling_mask_attribute_name {self.scaling_mask_attribute_name} not found in graph_object"
+        if self.scaling_mask_attribute_name not in self.nodes.node_attrs():
+            error_msg = f"{self.__class__.__module__}.{self.__class__.__name__}: "
+            error_msg += f"scaling_mask_attribute_name '{self.scaling_mask_attribute_name}' not found in graph_data - "
+            avail_masks = [k for k, v in self.nodes.items() if getattr(v, "dtype", None) == torch.bool]
+            error_msg += f"available boolean node attributes are: {avail_masks}"
             raise KeyError(error_msg)
 
     def reweight_attribute_values(self, values: torch.Tensor) -> torch.Tensor: