The removeEdge test could fail intermittently with the wrong order.
The precondition of a 1->2->3 order wasn't met, because there was no
edge from 1->3, so 3->1->2 was also a valid ordering.
The other failure was a bookkeeping error, were the recorded order may
not match the visited order. What happened in this case was the gateCh
was closed by V2, allowing V3 to run which could beat V2 to recording
its visit. Now the visit is recorded as part of the vertex walk, and the
gate is released as the final operation.
The order is deterministic now, so remove the brute-force test loop.
Because the vertex visit was record after the Update call, Updated
vertices may have been visited before the visit was recorded, causing
occasional test failures.
The order is now deterministic, and we can remove the brute-force loop.
