Removing rf edges when minimising may lead to missed executions

[johnwickerson]

If the execution involves the following thread

    a:W x 1 -po-> b:W x 2 -po-> c:R x 2

(plus some other threads to make the execution inconsistent), and b is removed, then we end up with

    a:W x 1 -po-> c:R x 0

which is certainly inconsistent. This means that the original execution will not be considered minimally inconsistent. Need to consider extending rf with co.rf in rf. Or else leave rf edges unconstrained after removal.