Fix ER sampling for undirected graphs

[smahdavi4]

Hi, I think there is a small issue in the implementation of ER sampling for undirected graphs. With current implementation, each edge is present if both of its directions are present, whose probability would be p^2. Also for self-loops, the probability is p. To make it more consistent with ER graph generation, I think each undirected edge should be sampled with probability p (instead of p^2).

[PetarV-]

Thank you for noting this!
For the next version of the library, we will decide if it is better to incorporate the pull request, or to leave a descriptive comment about the behaviour.