I don't think that lightlike is a crucial assumption; m=0 is just an approximation which does not affect the relevant physics.
I never thought about your question in detail; for me the structure of the matrix element simply followed from Lorentz covariannce which dictates that there are some trivial kinematical factors plus nontrivial functions involving F(x,Q²). The Lorentz structure of the matrix element is valid w/o ever referring to perturbation theory.
