Stress tensor (vc-relax) with KCP not implemented

[degennar]

Currently the stress tensor does not account for the possibility of complex wave functions.
The problem is that some of the involved quantities and routines do not have a complex "twin".

An example -- which causes the code crashing if one forces the calculation of the stress tensor with do_wf_cmplx=.true. -- is the calls to the routine dennl in any of the rhoofr_cp_* routines (see chargedensity.f90). In those calls to dennl one passes always bec%rvec, i.e. the real member of bec (type twin_matrix), also in the case of complex wfc calculations. For such calculation, only the complex member bec%cvec is defined while bec%rvec is not defined. One should then:

1. pass the proper member of bec (depending on whether bec%iscmplx is True or False)
2. adapt the routines receiving bec%rvec/cvec to deal with real or complex variables (i.e. creating twin versions of such routines in order to accept also complex values and creating the corresponding interfaces)

When do_wf_cmplx=.false., in principle the stress tensor is computed and one can perform a vc-relax calculation. However, it has not been tested whether the stress tensor is well computed for ODD functionals.