(chore) Update scalars_and_keras demo notebook to use learning_rate instead of lr (tensorflow#5363)

Removed deprecated lr and replaced with learning_rate. Tested the updated .ipynb file. [Here](https://colab.research.google.com/gist/jvishnuvardhan/25da1f6a06ddde2a7902e71c4aa404d4/scalars_and_keras.ipynb) is a gist for reference.

Author: jvishnuvardhan

docs/scalars_and_keras.ipynb

@@ -204,7 +204,7 @@
         "\n",
         "model.compile(\n",
         "    loss='mse', # keras.losses.mean_squared_error\n",
-        "    optimizer=keras.optimizers.SGD(lr=0.2),\n",
+        "    optimizer=keras.optimizers.SGD(learning_rate=0.2),\n",
         ")\n",
         "\n",
         "print(\"Training ... With default parameters, this takes less than 10 seconds.\")\n",