fix keras progbar output collision (2)

ensures that `keras.model.fit()` progbar output is complete before any output from custom callbacks appears on epoch end. Basically this edit puts the standard `ProgbarLogger` first in the list of callbacks. Originally discussed at [tf github](#43184) with problem description and gist examples. Latest example is [here]((https://colab.research.google.com/gist/poedator/3630b1cdec32ff6ef6ccdf6b63c4957a/custom_callback.ipynb).

tensorflow/python/keras/callbacks.py

@@ -264,7 +264,7 @@ class CallbackList(object):
 
     if self._progbar is None and add_progbar:
       self._progbar = ProgbarLogger(count_mode='steps')
-      self.callbacks.append(self._progbar)
+      self.callbacks.insert(0, self._progbar)
 
     if self._history is None and add_history:
       self._history = History()