Should a locale pack give a name's standard display and sorting forms? (parser_for(locales.ES) → sorting García López, Juan without writing the format) #1145

Split out of #754, which closed with this as its one leftover.

The question. A caller who wants a region's standard forms of a name has to know that region's convention and write it as a format spec:

>>> n = parser_for(locales.ES).parse("Juan García López")
>>> n.render("{family}, {given} {middle}")
'García López, Juan'
>>> n.render("sorting")
'sorting'

A Locale could carry the forms itself, in the shape of CLDR's person-name patterns (Unicode TR35 part 8): a given-first display pattern, a surname-first display pattern and a sorting pattern.

Decided (2026-10-28, decisions/vocabulary-model's #754 render-patterns entry):

• The shape: named patterns carried on the pack. The RFC's "Render patterns a pack may carry" maps Phase 4's views onto CLDR's fields.
• CLDR is where a pattern starts, not its authority. Its pt sorting pattern files under the first surname and drops surname2, which is not ABNT's last-surname entry (decisions/R2).
• The one gap the mapping showed, CLDR's per-surname prefix and core, is closed by surname1 and its views (decisions/R11, Should a family read as two surnames split into particles and base per surname? Juan Pérez de la Vega reads family_base 'Pérez Vega' #1022).
• The feature is additive: str() follows order and reads no pattern, and a pattern is something a caller asks for.

Open, the decision this issue is for: how a call reaches a pattern. ParsedName does not know the locale it was parsed under.

• (a) The parse records the convention on the name, as it records order. That costs one more ADDED_IN field, kept out of ==, and a new method or argument to ask for a pattern by. Reusing render()'s argument would change behavior, since render("sorting") returns 'sorting' today. The upside is that a name alone can give its forms.
• (b) Pattern rendering goes through the parser (Parser.render(name, "sorting") or the like), as Parser.capitalized does. That needs no field and raises no pickle question, but a name alone cannot give its convention's forms.

Also proposed, not decided: a sorting rendering is a string and sorts by code point. A key method beside comparison_key() could return the convention's fields in sort order, casefolded, for the caller to hand to real collation (ICU, locale.strxfrm).

Not in scope: gender, which decision 9 records only on readings.