gap orphans improvement : only lookup greatest existing ancestor, then cache

Follow-up of OAK-10761 : to further improve gap vs non-gap orphan type detection, lookup only the direct child of the greatest existing ancestor, then cache that result. Upon further type detection, use that cache to start with. That eliminates the tree traversal, plus using the cache should further limit the number of lookups in the first place. The cache can be small (eg 64) and assuming paths aren't excessive (say max 10k) the resulting memory usage would still be small (eg 64k characters).