Planner: simulated annealing should work for planning problems with hard constraints often broken too

This code is bad in SimulatedAnnealingAcceptor:

// TODO don't abuse translateScoreToGraphValue
// TODO do hard and soft separately and then average their acceptChance
Double diff = localSearchSolverScope.getScoreDefinition().translateScoreToGraphValue(scoreDifference);
if (diff == null)

{ // more hard constraints broken, ignore it for now return 0.0; }