Instead of midpoint density, propagate the whole timestep and
guess the final density before the self-consistent iterations start.

Some handling of matrices could have been a bit confusing in the
apply_delta_pulse subroutine - this has now been harmonized with the
rest of the code.

As more options specialized for the RT-BSE scheme emerge (with still
more on the horizon) a specialized input section is designated,
which handles the options separately from TDDFT options

The regularization parameter set in `gw_utils.F` was set only for the
original limit of size for the minimax grid, which was 20. However,
additional larger grids are now available, but remain unregularized - no
small number is added to stabilize the singular value decomposition.

The effect is observed by difference in the reported correlation
self-energy for different number of MPI ranks in an otherwise identical
setup.

This change allows the same regularization procedure to be applicable to
all grids of size > 20.

A TDDFT run does not use certain print sections, while RTBSE run does
not use certain other sections. In order to not confuse the user about
which section is available for which code, explicitly print out warning
message when an unused print section is present.

So that they can be called from other places in the code

get_sigma can now in principle be called without the reference to
rtbse_env

Added conditions that require screened potential to be calculated even
when starting from initial energy run.

On some architectures, it might be better to run the GW calculation
beforehand and RTBSE calculation afterwards. This patch simplifies the
rerun procedure.