Unable to specify custom role for Sagemaker

# Issue Description
Unable to use Sagemaker role when provisioning Sagemaker resources. Here's how I am used to do it through AWS Console:


Found this in Sagemaker Python library docs:


Tried leveraging with:
```
backend_args = {
"autogluon_sagemaker_estimator_kwargs": {
"role": "arn:aws:iam::176253007097:role/service-role/SageMaker-winnie-ml",
}
}

cloud_predictor.fit(
job_name="insights-autogluon-refund-ml",
predictor_init_args=predictor_init_args,
predictor_fit_args=predictor_fit_args,
backend_kwargs=backend_args,
)
```

But got this error:

`TypeError: autogluon.cloud.utils.ag_sagemaker.AutoGluonSagemakerEstimator() got multiple values for keyword argument 'role'`