There seems to be a misunderstanding here. My comment was not meant as an response to your feature request in general, but solely meant as an answer to your "why" question.
Of course, there are ways to build neighborhood views other than the one we have chosen. (Case in point, the one you suggest in your original post.) And, of course, it makes sense to request different behaviors. Actually, we do appreciate such suggestions and feature requests very much.
However, that does not mean the current behavior is broken. Indeed, "In this view a missing edge is not consistent with a definition of shortest path, the path is corrupted due to missing edge." is just plain wrong. Solutions for *destination* shortest-path problems never contain edges between source nodes. Moreover, the statement "In this point of view a missing edge to neighbor (regardless if the neighbor is also selected or not) is a bug." is just not applicable, because the behavior you would like to have is not the behavior that is currently implemented.
Again, I am not saying your request is not valid. All I am saying is that the current behavior is correct with regards to the specification that was originally decided on.