Add TF CIFAR test to continuous testing set (#170)

Though this test has historically been relatively flaky, there was
a recent change to the training steps that helped stabilize it, so
it should be fine to include this test with our continuous testing.
This test is important because it's currently our only one that tests
distributed training and the only one to run on GPU instances.

Author: laurenyu

tests/integ/test_tf_cifar.py

@@ -14,6 +14,7 @@
 import pickle
 
 import numpy as np
+import pytest
 
 from sagemaker.tensorflow import TensorFlow
 from tests.integ import DATA_DIR
@@ -30,6 +31,7 @@ def __call__(self, data):
         return pickle.dumps(data, protocol=2)
 
 
+@pytest.mark.continuous_testing
 def test_cifar(sagemaker_session, tf_full_version):
     with timeout(minutes=20):
         script_path = os.path.join(DATA_DIR, 'cifar_10', 'source')