In CalculationData, the bfield is read for every atom and then initialized
only for the locally treated atoms. Thus, the bfield has to be accessed
differently from the other quantities that are only allocated for the
local atoms.

Also, I restructured and commented lots of code while looking for the bug.

The data is now stored at appropriate places and loaded in the
corresponding part of the code.
It is not used for any calculation yet.

Code compiles, but is not tested in any other way.

The parameters are taken from Sascha Brinkers presentation and should
match what is implemented in KKRhost.
The names are kept as close as possible to the inputcard keywords in
KKRhost, thus they might differ from what they are called in the
KKRhost code.

Added the parameters to datastructures/InputParamsNew.txt and
generated a new datastructures/InputParams_mod.F90 using
the generate.sh skript.

In generate.sh, the inputgenerator.py is now called as a local script.

In inputgenerator.py, use /usr/bin/env to be independent of actual python installation dir.
Explicitly use python2, which is needed by the skript and not always the default anymore.

The parameter enable_quad_prec was introduced in InputParams_mod.F90,
but not in InputParamsNew.txt, so it did not appear when recreating
the Fortran module from the list of parameters.

mkdir -p build
make tfQMRgpu SMP=openmp tfQMRgpu=yes
make SMP=openmp tfQMRgpu=yes -j

Adopted MakeFile, swiched off grouping and changed transposition flags for gpu call.

Adds in input possible specification of the a_log parameter of the exponential mesh. Removes hard-coded a_log.

reintroduced tolerance value 1.d-16 in function half-space, seems to be necessary for large-scale MnGe B20 system

- NOTE: ZnO regtests were deactivated because of problems with the atomic core solver (SHOULD BE FIXED!)