h- and r-adaptivity.

A:
* h-adaptivity gets worse with increasing p (not correct)
* GH: choose nIter=1 ==> all solutions are incorrect
* GH: choose nIter=20 ==> all solutions are correct
* GH: nIter = 5 and Dirchlet bc at x=1.5 ==> aweful for p=20, 100
  Integration of f(x) is not accurate enough. ==> better integration formula needed, see integral(...)
  
B:
* h-adaptivity is correct

C:
* r-adaptivity
  looks suspicious, but similar to Mandl
  cumsum in r_adapt() not correct, but exact cumsum results in wrong solution behavior