added GAT and GCN message passing layers

[oreo-hippiepham1]

added implementations for GAT (this was in the 2023 colab, however was not implemented in this repo yet. Other available implementations seem to be wrong, so I did my own version) + GCN (mostly following PyG's MessagePassing tutorial)

[karannb]

Hi,

• the GCN layer's implementation is perfect, but it would be great if you could run and show the train-test stuff with all the models; and cite some source where you have taken the math from (was it https://towardsdatascience.com/graph-convolutional-networks-introduction-to-gnns-24b3f60d6c95 ?).
• However I think the GAT paper (https://arxiv.org/pdf/1710.10903.pdf) had only one feed-forward for both source and destination (your code has two, attn_l and attn_r) and similarly it would be better if you run this model's train and evaluation as well.
• You have provided great comments!
  Thank you.