You can't do this with the hybrid potentials as they are currently
setup in LAMMPS. You could specify C C C C for types 1-4 with
the pair_coeff command which would do what you want except
that you will also get Tersoff interactions between the 2 CNT,
assuming they are close enough to be within cutoffs. You could
also superpose on top of this (via hybrid/overlay) some additional
pair potential for interactions between atoms in 2 different CNT (e.g. LJ).
Physically, I can't imagine a potential (like Tersoff) that interacts
only between atoms in one CNT but magically ignores nearby C atoms
that are in another CNT. That's not what a bond-order potential
like Tersoff means conceptually.