Remove velocity from partially frozen atoms in md-vv
md-vv would add some velocity to the frozen dimensions of partially
frozen atoms during constraining. This did not lead to wrong
trajectories, as the frozen dimensions of the positions are kept fixed
during propagation. The non-zero velocities were, however, reported in
trajectories and final configurations. They might also have lead to
slightly wrong kinetic energies, since the reported kinetic energy is
calculated after the velocities are constrained. All effects are
expected to be relatively small, since they did not accumulate, as the
velocities were regularly reset to zero once per step.
Refs #3849