Remove use_language_model param from process_batch

No need for use_language_model param since we rely on the model's lm_decoder anyway.