specify -DCPP_OLDINTEL to deactivate parts of the code that did
not compile; mainly the order of USE statements was changed or
ONLY was added to avoid confusion of the compiler.

Removed dimension%nn3d and dimension%nn2d variable and replaced them with the actual dimensions kimax+1,kimax2+1

Added explicit dimensions in calls to xcpot-routines to make code more stable. Fixed bug that should fix #191.

Bugfix in XC potential in MT spheres in cases with many atoms different jri and jspin=2. Should fix #193 @mhoff, please check.

   ...vXC%pw_w was not initialized.

   + some code beautification

I already commented this out in commit 415885af.
I don't know who commented it in again, but the problem with this parallelization is
not solved. With ifort 18 I obtain for the CuBulkXML test with this parallelization
a qlmp(0,0,1) of (1.490549732270697E+307,-2.695127599127286E-033). Without the
parallelization it looks okay: (0.990659014383078,-2.695127599127286E-033). It may
be that only the 0,0,1 element is affected. But this should be checked before the
parallelization is commented in again.

The preconditioner already improves the convergence of Cu and Fe files, but the starting distance is higher than for the older fleur version and tests run up to test 7 only (does not fail, but hangs).