Some time ago I modified routine at2celldm so that it computed the value of

ibrav even if it was set to 0 and lattice defined by at vectors. It wasn't
such a great idea: may break the phonon code, produces a different output
for scf and non-scf runs, does not account for the unlikely but not impossible
case of rotatted cell. This commit reverts to the previous behaviour, while
still allowing to guess an ibrav if desired.