That's good. About standard alternatives, REFN could fit the bill:
A user-defined number or text that the submitter uses to identify this record. For instance, it may be a record number within the submitter's automated or manual system, or it may be a page and position number on a pedigree chart.
It is multivalued and takes a TYPE subtag that can make it unique. Unfortunately, no one uses it like that, preferring the non-standard _UID instead. A problem is that no definition of _UID exists that I could find. Apparently, it was introduced in PAF
5.1 in 2002 taken from an earlier draft, that I cannot find. So it is unclear if it supports my understanding of it.
But, anyway, let's separate the case for GRAMPS storing multivalued unique identifiers and using them for matching purposes, from the specific mappings on import and export for different formats. The former I think is a must, for the latter I grant there is a lot of room for personal taste.