Detection of cycles too defensive

[meistermeier]

With a graph like this

the internal logic will determine C->D as an indicator for a cycle.
It gets processed twice by A->B->C->D and A->C->D

This should be a "directed" query instead of the data-driven one for cycles.
BUT I don't know yet if it makes sense because with a "directed" query, we would duplicate all the nodes and relationships behind the D in two separated branches starting from A.