Implement Global attention mechanism

[akhilpandey95]

resources to consider

• https://pytorch-geometric.readthedocs.io/en/1.3.1/_modules/torch_geometric/nn/glob/attention.html

[allaffa]

@akhilpandey95

Global attention can mean many things. A few examples are:

• concatenate global input features to the graph embedding layer created by the message passing layers.
• a "slight" generalization of the above consists in passing the global input features into an MLP, and then concatenate the output to the graph embedding generated by the MPNN layers