Investigate whether we can calculate better PairPos format1 subtable splits by tracking shared subtables

This is fairly low priority, since the cost of an additional subtable is low, and this particular optimization is unlikely to save us more than a split or two, but I'm curious anyway.