Ugly but fast. :)

For the serial case there was a timer disbalance in some related subroutines.
This looks systematic and may also be present in other Wannier subroutines
that are not covered by the test.

Will break stuff.

# Conflicts:
#   eigen/eigen.F90
#   init/step_function.F90
#   vgen/mpmom.F90
#   vgen/vmts.F90

This also re-enables the OMP parallelization in this subroutine.

The calculation of the number of OMP threads was broken.

The loop structure is strongly refactored. It is not yet clear whether
it fixes issue #672. This still has to be tested. The OMP reduction
in the former approach has been refactored away.