ml-frameworks/tensorflow/train-tensorflow-resume-training uses http://yann.lecun.com/exdb/mnist and tends to fail

I'm using https://github.com/Azure/MachineLearningNotebooks/tree/6f893ff0b40d3d6af94ac43867676a3d9df1fcbd/how-to-use-azureml/ml-frameworks/tensorflow/train-tensorflow-resume-training on a default compute instance with "Python 3.6 AzureML" kernel.

```sh
find . -type f -name "*.ipynb" -or -name "*.py" | xargs grep -m 1 "lecun.*gz" | cut -d: -f1
./how-to-use-azureml/deployment/onnx/onnx-train-pytorch-aml-deploy-mnist.ipynb
./how-to-use-azureml/ml-frameworks/tensorflow/train-tensorflow-resume-training/train-tensorflow-resume-training.ipynb
```

Related to https://github.com/pytorch/pytorch/issues/53267.

Use "http://yann.lecun.com/exdb/mnist" -> "https://azureopendatastorage.azurefd.net/mnist" instead.