Exported Sklearn Prediction script fails when the configured ground truth column is missing from the input #8864

What happened?

A workflow with a Sklearn Prediction operator runs fine in Texera when the "Ground Truth Attribute Name to Ignore" names a column missing from the data input. The engine silently ignores the missing name and scores the run. Exporting the same workflow as a standalone Python script and running it fails instead: the script exits with a KeyError naming the missing column and writes no predictions. I expected the exported script to behave like the engine and complete the run.

How to reproduce?

1. Build a workflow with a Sklearn Prediction operator fed by a fitted model frame and a data frame.
2. In the operator's properties, set "Ground Truth Attribute Name to Ignore" to a column name that does not exist in the data frame. A typo or a stale name from an earlier schema will do.
3. Run the workflow in Texera: it completes and produces predictions.
4. Export the workflow as a Python script and run it: the script ends with a KeyError for the missing column.

Version/Branch

1.4.0-incubating-SNAPSHOT (main)

Commit Hash (Optional)

4f36c15, verified still present on origin/main.

Relevant log output

The exported script exits with a KeyError naming the configured ground truth column as not found in the axis, and no output predictions are produced.