when storing computed lightmap on each bounce, user can interrupt compilation and
get working files, but it spends allocation time (which is slow and single-threaded

with this option, user can decide to only allocate lightmaps on the very final
bounce, it means it can't be interrupted, but it can save a lot of time