Save all stage 1 sum factorization kernels that are used in
accumulation expression in the cache during the dry run. Discard all
inactive sum factorization kernels in decide_vetorization_strategy.