[rules-dev] shadow facts

If we know that facts will only have their values altered in the consequence, during the execution of the working memory, we can dispense with shadow facts - instead we can alter the "setter" to do a retract and assert. What was the overhead of shadow facts in the end, about 10%? so this should help the benchmarks. This should also help with problems such as marcos's, where he is dealing with a million and upwards facts. Mark