Trying to fix poor interaction between control flow and tpu estimator.

Control flow v1 gradients depend on control flow contexts implementing AddValue.

PiperOrigin-RevId: 314162889

Author: alextp

tensorflow_estimator/python/estimator/tpu/tpu_estimator.py

@@ -3834,6 +3834,10 @@ def AddOp(self, op):  # pylint: disable=invalid-name
         raise ValueError('{}: Op {} depends on TPU computation {}, '
                          'which is not allowed.'.format(self._message, op, c))
 
+  def AddValue(self, value):
+    self.AddOp(value.op)
+    return value
+
   def __enter__(self):
     # pylint: disable=protected-access
     self._g = tf.compat.v1.get_default_graph()