Add more example scripts

[abieler]

• rename examples/gcn.jl --> examples/gcn_gpu.jl. (this actually does not run, but IDK if its my setup or a bug)
• added examples/gcn.jl (original example but w/o the |> gpu parts
• added examples/gcn_featured_graph.jl (was a little more tedious to figure out how to make this work than i hoped from the docs alone)
• note Dropout is broken for the FeaturedGraph example

[CarloLucibello]

rename examples/gcn.jl --> examples/gcn_gpu.jl. (this actually does not run, but IDK if its my setup or a bug)
added examples/gcn.jl (original example but w/o the |> gpu parts

this seems a wasteful code duplication. You could just add a use_cuda flag, see e.g. https://github.com/FluxML/model-zoo/blob/master/vision/mlp_mnist/mlp_mnist.jl

[CarloLucibello]

what's the error?

[CarloLucibello]

Suggested change

	accuracy(x, y) = mean(onecold(softmax(cpu(model(x)))) .== onecold(cpu(y)))
	accuracy(x, y) = mean(onecold(cpu(model(x))) .== onecold(cpu(y)))

[CarloLucibello]

onecold doesn't need normalized predictions

[yuehhua]

For GCNConv, if you pass it a FeaturedGraph, it will return you a FeaturedGraph. If you pass it an array, it will give you an array back. So, in the context of input being a FeaturedGraph, a Flux layer should not being used in the network, or you should pass it by bypass_graph.