Remove upranking behavior from normalization

We would like normalization to be usable for unbatched multidimensional samples data in a dataset.
This removes the automatic upranking of rank 0 and rank 1 inputs so this is possible.

As a consequence, the default axis=-1 will treat rank 1 inputs as single samples, and
normalize each dimension of the input individually.

If you would like to pass batched or unbatched scalar data into the layer, and normalize
every element, you should pass axis=None.

PiperOrigin-RevId: 379868721

Author: tensorflower-gardener

site/en/tutorials/structured_data/preprocessing_layers.ipynb

@@ -385,7 +385,7 @@
       "source": [
         "def get_normalization_layer(name, dataset):\n",
         "  # Create a Normalization layer for our feature.\n",
-        "  normalizer = preprocessing.Normalization()\n",
+        "  normalizer = preprocessing.Normalization(axis=None)\n",
         "\n",
         "  # Prepare a Dataset that only yields our feature.\n",
         "  feature_ds = dataset.map(lambda x, y: x[name])\n",