Adds support of undefined (None) number of vertices/faces/colors in the mesh.

How to test:

• Run updated unit tests.
• Modify mesh_demo.py to provide different number of points at each run:

# Add extra vertex (1,2,3) to the mesh.
vertices_2 = np.concatenate((vertices, [[1, 2, 3]]))  
vertices_2 = np.expand_dims(vertices_2, 0)
...
for i in range(_MAX_STEPS):
    ver = vertices if i % 2 == 0 else vertices_2
    summary = sess.run(meshes_summary, feed_dict={
        vertices_tensor: ver,
        faces_tensor: faces,
        colors_tensor: colors,
        step: i,
    })

Run demo app and then launch tensorboard - observe mesh being rendered.

Also this PR changes the way we load data - from now on, every step is considered as unique data point and should be requested separately from the server (similar to image plugin). Please note that each mesh/step will produce 1 to 3 requests based on what data was passed into the summary (i.e. if there are no colors specified, then it will request only vertices and faces).

This PR depends on some TF-Graphics changes (I'll update it with new SHA once changes merged).