Output per atom energy for a specific pair

Probably can use compute/group/group. I know you can get the per-atom
energy by interacting a group of one atom with group all. You would
just need to change the groups to only include LJ interactions. You
would also have to repeat for each atom you want to get per-atom
energy for, so it wouldn't be very efficient... I think it is doable
though. Not sure if it would just be easier to write a new compute.