This uses the foreach_set_index() method introduced in the previous
patch to speed up the copy bitvectors in ACD nodes, as pointed in
issue #476.
Running
PREFIXCMD='valgrind --tool=callgrind' ./run python3 -c \
"import spot; spot.acd_transform(spot.automaton('syntcomp_91.hoa'))"
went from 65139436227 instruction fetches down to 18490399159.
* spot/twaalgos/zlktree.cc (acd::build_): Use foreach_set_index().