Enable external predictions for short model in benchmark

[SvenKlaassen]

Description

Contributed by @lucien1011.

This pull request adds an optional input arguments fit_args to the function sensitivity_benchmark in the class DoubleML. Most importantly, this addition will enable the usage of external_predictions when fitting short models for sensitivity analysis.

The new argument has to be in a nested dictionary like this:

dataset = dml.DoubleMLData(
    df,
    y_col='y',
    d_col='d',
    x_cols=cov_cols,
    force_all_x_finite=False,
)

dml_irm = dml.DoubleMLIRM(
    dataset,
    ml_g=RandomForestRegressor(), #dummy learner only
    ml_l=RandomForestClassifier(),
)

# Some user-specific codes to calculate external_predictions and put in the following columns
# df['d_prop'] df['y_pred_d0'] df['y_pred_d1']

bm = dml_irm.sensitivity_benchmark(
    benchmarking_set=['covariate_to_be_tested'],
    fit_args=dict(
        external_predictions=(
            d=dict(
                ml_m=df['d_prop'].to_numpy().reshape(-1,1),
                ml_g0=df['y_pred_d0'].to_numpy().reshape(-1,1),
                ml_g1=df['y_pred_d1'].to_numpy().reshape(-1,1),
            ),
        ),
    ),
)

Reference to Issues or PRs

No related issues or PRs to my knowledge.

PR Checklist

Please fill out this PR checklist (see our contributing guidelines for details).

• The title of the pull request summarizes the changes made.
• The PR contains a detailed description of all changes and additions.
• References to related issues or PRs are added.
• The code passes all (unit) tests.
• Enhancements or new feature are equipped with unit tests.
• The changes adhere to the PEP8 standards.