Description

• Replaced the removed tensorflow.examples.tutorials.mnist.input_data dependency with a small MNIST loader backed by tf.keras.datasets.mnist.
• Updated the affected TensorFlow tutorial scripts and notebooks to use tensorflow.compat.v1 for TF1-style graph/session APIs.
• Updated saved-model and graph-freezing calls to use TF1 compatibility paths.

Motivation and Context

• Newer TensorFlow versions no longer include tensorflow.examples, which caused ModuleNotFoundError: No module named 'tensorflow.examples' when running the TensorFlow-to-ONNX MNIST tutorials.
• The tutorials still use TensorFlow 1.x APIs such as sessions, placeholders, reset_default_graph, and graph freezing, so routing those calls through tensorflow.compat.v1 keeps the original tutorial flow working on modern TensorFlow installs.
• Fixes the reported notebook failure during step 1 model creation/training.