…lator

In dfpt_nsteltwf (m_dfpt_scfcv.F90) the slices that copy ground-state
and first-order wavefunctions out of cg/cg1 mixed the global band index
(iband) on the upper bound with the local-to-MPI-process index
(iband_me) on the lower bound:

  cwave0(:,:)=cg(:,1+(iband_me-1)*npw_k*nspinor+icg:iband*npw_k*nspinor+icg)
  cwavef(:,:)=cg1(:,1+(iband_me-1)*npw1_k*nspinor+icg1:iband*npw1_k*nspinor+icg1)

Whenever DFPT band parallelism is active (npband > 1, which ABINIT
auto-selects in many strain-response runs even with paral_kgb = 0),
iband > iband_me on every rank that does not own the lowest band group,
so the slice is longer than the destination buffer (npw_k*nspinor) and
overruns cg/cg1. This produces a SIGSEGV inside dfpt_nsteltwf during the
strain (rfstrs) dataset of a 5-dataset DFPT pipeline; the crash was
observed in a 200-atom MOF DFPT run on ABINIT 9.8.2 and reproduces on
current master (v10.6.7).

Fix: use iband_me consistently on both bounds, so the slice length is
always exactly npw_k*nspinor (resp. npw1_k*nspinor), matching the
allocation of cwave0/cwavef.

Related to the issue describing this segfault and reproducer.