-h to enable:
split rules into terminal sequences, look up each in the suffix array,
and treat the intersection of the resulting sentence lists
as occurrences of the rule.
for a speedup, use unordered_set instead of set for SentIdSet
(requires GCC >= 4.3)