This sounds pretty good to me. Bit of a flexible use of DWARF, probably want to propose it for standardization, perhaps - but sounds pretty plausible.
& you make a good point about linked/merged indexes - even if we could use some hint about whether a single index has parents or doesn’t (so if it has parents, the absence of a parent means the node has no parent, not that the parent information has been omitted) - it wouldn’t apply to a merged index. So we need a per-entry parent-absent disambiguator.