gamut-iptc: an extension field the projection cannot express reads as absent — should the model widen?

[justin13888]

gamut_iptc::extension now decides retention by round-trippability: a field enters the typed
value only when the property the writer will emit for that value reproduces the field that was read
— value, RDF container kind, qualifiers and all. Everything else is kept verbatim in the type's
other list (PR #540, round 4). Nothing is lost from the graph any more.

The cost is on the read side. A field the projection cannot express now reads as None (or an
empty Vec) where it previously returned a value the writer would then have destroyed. The module
enumerates the shapes; the ones a real file is most likely to carry are:

• a URL held as rdf:resource rather than as element text — Iptc4xmpCore:CiUrlWork,
  Iptc4xmpExt:AOSourceInvURL, plus:LicensorURL, and any item of xmp:Identifier;
• a value carrying a qualifier, most often an xml:lang on a plain text field;
• a language alternative holding a language beside x-default;
• an rdf:Seq where the model writes an rdf:Bag, or the reverse.

The question this issue holds open: should the model grow to express them, so the typed accessor
reports the value and the writer puts it back as it found it?

Sketch of the fork:

• Leave it. The graph is intact and the field is reachable through other. Cheapest, and the
  typed view stays plain data (Option<String>, Vec<String>) as AGENTS.md's C-portability
  convention wants. A caller reading licensors()[0].web_url on a PLUS-conformant file written
  with rdf:resource gets None and has to know to look in other.
• Carry the value kind. Replace the URL fields' String with a type that remembers whether it
  was rdf:resource. Breaking for three public fields, and Entity::identifiers is a Vec<String>
  where the kind is per item, so it needs a per-item type — a heavier public API for what is
  otherwise plain data.
• Carry the whole language alternative. Option<String> becomes a list of (tag, text), which
  is the honest model of an rdf:Alt but changes the ergonomics of every descriptive field.
• Carry qualifiers. Decision 4 of round 4 rejected this as larger than gamut-iptc: IPTC Extension structures + IIM records 3-9 #422's scope; it would
  still leave the general case open, because a qualifier can hang off anything.

Whichever way it goes, it should be decided once for the whole extension module rather than per
field, and the answer interacts with #591 (whether other should be public at all) and #573
(whether emission should be fallible).

Scope: crates/gamut-iptc/src/extension.rs. Refs #422, and PR #540, where the retention rule and
its enumeration landed.