[Frontend][Tensorflow] Fix TF 1.15 conv2d_transpose parsing

[jtuyls]

Fix for Tensorflow 1.15.0 conv2d_transpose parsing. Specifically, attr["_output_shapes"][0] can be None.

Following snippet can be used to create a model for reproducing the bug (run with TF 1.15.0):

import tensorflow as tf

output_shape = [1, 8, 8, 128]
strides = [1, 2, 2, 1]

x = tf.compat.v1.placeholder(tf.float32, shape=[1, 32, 32, 4])
w = tf.constant(0.1, shape=[7, 7, 128, 4])
ct = tf.nn.conv2d_transpose(x, w, output_shape=output_shape, strides=strides, padding='SAME')

graph_def = tf.compat.v1.get_default_graph().as_graph_def()
with tf.io.gfile.GFile('conv2d_transpose_tf1_15.pb', "wb") as f:
    f.write(graph_def.SerializeToString())

In general, I think it might be useful to adjust the CI to run the frontend tests with multiple versions of the the external frameworks to verify compatibility with the versions we want to support?

@siju-samuel @masahi, could you have a look?

[siju-samuel]

@jtuyls Good catch. Thanks for the contribution.
Can you add a testcase convering this scenario as well?

[jtuyls]

@siju-samuel I guess this issue is really caused by not using AddShapesToGraphDef but I think it's still useful to include the small changes to cover this case. So I added checks for that to the Tensorflow convolution tests. Could you have a look?

Additionally, we could maybe add tests for this case to the other operations?

[siju-samuel]

LGTM

[tqchen]

Thanks @jtuyls @siju-samuel