Network Isolation for TensorFlow Training Requires Setting `model_dir` to `False` when creating a TensorFlow Estimator Object

[evakravi]

Describe the bug
In order to enable network isolation for Tensorflow training, we must set model_dir=False when creating a TensorFlow Estimator object. See:

sagemaker-python-sdk/src/sagemaker/tensorflow/estimator.py

Lines 69 to 70 in 9b437be

	To disable having ``model_dir`` passed to your training script,
	set ``model_dir=False``.

This is not intuitive for the user, and took several months for SageMaker JumpStart team to identify.