changed all atom_vec variants to have the forces (and torque) arrays to be allocated to nthreads times the usual size. this will allow to have cocurrent threads compute forces (and torques) for maximum efficiency. non-threaded lammps compiles are not affected (nthreads defaults to 1, ignored as input command in non-OpenMP compiles and otherwise guaranteed to be >0). this small change doesn't warrant writing new atom styles...