-In triqs version 2.2 array_views no longer own the memory they point to
This means that array variables that are local to a function should always
be returned as arrray and never as an array_view
In order to wrap the ATM routines by Python using TRIQS wrapping
tools it is necessary to modify the interface to 'dos_tetra3d'.
The major changes involved replacing direct NumPy arrays with
TRIQS arrays which can be converted to Python arrays using library
tools.
Also, some small changes were necessary to port the functions
from C99 complex numbers to C++ style.
CMakeList is added to automatize building of the ATM library.
