Convolutions on heterogeneous graphs

[salbert83]

I noticed 3 things: 1) HeteroGraphConv is not available in GNNLux (although available in GraphNeuralNetworks.jl). Any reason? 2) HeteroGraphConv considers the subgraphs corresponding to the edge types and performs graph convolutions on these. There is an implicit assumption that the source and target data will have the same dimension, which is almost never the case for a heterogenous graph. Finally, 3) I don't see a convenient way of incorporating edge data in the convolutions for heterogenous graphs. The current implementation just extracts node data. I know I could implement custom layers using message passing, but it seems to me these are fairly generic things one would expect. Am I missing something? Thanks in advance and apologies if I misunderstood the current state of the code.