This makes the methods AverageDistanceAssembler::assembleHessian
compile again.  And it should be a wee bit faster two, because
matrix symmetry is now exploited better.