[BUG] Not implemented decision_function method in ExtendedDataFramePipeline

[samihamdan]

Describe the bug
When using scoring in run_cross_validation which need a decision function (e.g. roc_auc) we run into a bug.
The problem is that we did not implement a decision_function in the ExtendedDataFramePipeline.

To Reproduce

from seaborn import load_dataset
from julearn import run_cross_validation


df_iris = (load_dataset('iris')
           .query('species in ["versicolor", "virginica"]')
          )

X = ['sepal_length', 'sepal_width', 'petal_length']
y = 'species'
scores = run_cross_validation(
    X=X, y=y, data=df_iris, model='svm',scoring='roc_auc')

will throw:
AttributeError: 'ExtendedDataFramePipeline' object has no attribute 'decision_function'

Expected behavior
As the normal cross_validate function we should be able to use this scoring method.
The ExtendedDataFramePipeline should just use the decision_function from the used model (here svm).

Screenshots

System (please complete the following information):

• OS: Linux 10/ Debian
• Desktop Gnome 3.30.2

Additional context
Should be easy to fix.