N in this case is low, as it's the number of distinct projects referenced from
MRs, rather than the number of MRs referenced (issues use their own
optimization). Still, on issues or MRs which are often referenced from MRs, it
will save a few queries.