Effect modification in a randomized trial under non‐ignorable non‐compliance: an application to the alpha‐tocopherol beta‐carotene study