Topology optimization: more robust

[MaxVZIKM]

It was observed, that in the code example for topology optimization the update of the densities only depends on \eta. Correct is to multiply eta with the iterations needed for stability n_j. Otherwise, a behavior similar to checkerboarding may be observed for larger 3D problems. The density simply is updated too fast. Thus, with this issue the example is more robust.