Should undirected edges == and hash be undirected?

Edges for SimpleGraph still consider their directedness when testing for equality and hashing. This is understandable given that SimpleEdge is used for directed graphs also. Is it worth considering adding a SimpleUndirectedEdge (or similar)? Something along the lines of:

...

function ==(e1::SimpleUndirectedEdge, e2::SimpleUndirectedEdge)
    return (src(e1) == src(e2) && dst(e1) == dst(e2)) ||
        (src(e1) == dst(e2) && dst(e1) == src(e2))
end

hash(e::SimpleUndirectedEdge, h::UInt) = hash(src(e), h) ⊻ hash(dst(e), h)

And then having SimpleGraph instead return edges of this type by default?

This would more easily allow users to maintain sets / dictionaries of edges. For instance, the reason I'm opening this is the associated awkwardness I encountered when implementing a minimum cycle basis that maintains such edge sets.

Happy to do a PR if this change is palatable.

Start by reviewing SimpleGraph's edge return behavior alongside SimpleEdge equality and hash methods. Compare the proposed SimpleUndirectedEdge behavior with the minimum cycle basis use case and existing edge-set expectations. Done means the project has an agreed direction for undirected edge identity and a corresponding, validated API change.