Add more convenience constructors

The type GNNGraph from GraphNeuralNetworks.jl is a Graphs.AbstractGraphs and implements the corresponding interface, therefore it would be nice to make the following code work by providing an appropriate constructor:

julia> using GraphNeuralNetworks, Graphs

julia> g = GraphNeuralNetworks.rand_graph(10, 20)
GNNGraph:
    num_nodes = 10
    num_edges = 20

julia> g isa Graphs.AbstractGraph
true

julia> Graphs.SimpleGraph(g)
ERROR: MethodError: no method matching SimpleGraph(::GNNGraph{Tuple{Vector{Int64}, Vector{Int64}, Nothing}})
Closest candidates are:
  SimpleGraph(::Any, ::Array{Vector{T}, 1}) where T at ~/.julia/packages/Graphs/pxGmw/src/SimpleGraphs/simplegraph.jl:18
  SimpleGraph() at ~/.julia/packages/Graphs/pxGmw/src/SimpleGraphs/simplegraph.jl:50
  SimpleGraph(::SimpleGraph) at ~/.julia/packages/Graphs/pxGmw/src/SimpleGraphs/simplegraph.jl:115
  ...

Moreover, we could also have the following constructor:

julia> src, dest = edge_index(g)
([1, 1, 1, 1, 2, 2, 2, 5, 5, 5, 2, 3, 7, 10, 5, 6, 7, 6, 7, 9], [2, 3, 7, 10, 5, 6, 7, 6, 7, 9, 1, 1, 1, 1, 2, 2, 2, 5, 5, 5])

julia> SimpleGraph(src, dest)
ERROR: MethodError: no method matching SimpleGraph(::Vector{Int64}, ::Vector{Int64})
Closest candidates are:
  SimpleGraph(::Any, ::Array{Vector{T}, 1}) where T at ~/.julia/packages/Graphs/pxGmw/src/SimpleGraphs/simplegraph.jl:18

which is more concise and discoverable than

julia> SimpleGraph(Edge.(src,dest))
{10, 10} undirected simple Int64 graph

Start by tracing the existing GNNGraph, rand_graph, edge_index, and Graphs.SimpleGraph entry points, then inspect how Graphs currently constructs graphs from edges. Use the REPL examples in the issue as acceptance checks: conversion from GNNGraph and construction from source and destination vectors should work, while preserving the existing Edge-based form.