Extracting molecular coordinates for QM9 dataset from sdf files

[arunppsg]

The .sdf files comes with information about molecular coordinates. The molecular coordinates can be used in wide range of predictive models for predicting molecular properties. This PR updates the existing load_sdf_files function to extract molecular coordinates from .sdf file. Relevant changes have been made to MolGraphConvFeaturizer to support molecular coordinates as kwargs.

Brief demonstration of the change: in the following code,

tasks, datasets, transformers = dc.molnet.load_qm9(reload=False, featurizer  =dc.feat.MolGraphConvFeaturizer())
train = datasets[0]
train.X[0]

the current output is
GraphData(node_features=[17, 30], edge_index=[2, 34], edge_features=None, pos=[17, 3]) whereas, earlier it was GraphData(node_features=[17, 30], edge_index=[2, 34], edge_features=None)

Additionally, tests have been added for load_sdf_file and MolGraphconvFeaturizer to capture the new features, renamed test data directory to assets from data and made yapf fixes.

[rbharath]

I took a first look, but it might be easiest to review the PR on the developer call or office hours since I want to make sure I understand all the changes. Better 3D coordinate handling in general seems like a worthwhile change! Just want to make sure I understand the details

[rbharath]

Are these standard names for the position columns or are they specific to the qm9 dataset?