messages: slice-3 ICU MessageFormat exclusions (selectordinal, offset:, =N, skeletons, nested dispatch) — build on demand

[accuser]

Summary

Tracking issue for the ICU MessageFormat constructs slice 3 (#878, ADR 0276) deliberately excluded. Each is diagnosed today via bynk.messages.malformed_icu_syntax (not silently mishandled), so authoring one gives a clear "not supported" error rather than wrong output. This issue enumerates them so the deferral is on the tracker rather than only in the ADR; each is a separable increment to open a proposal for when real demand appears.

Not planned as a batch — the surface was capped on purpose (ADR 0276). File this so nothing is lost, not to schedule it.

The excluded constructs

Construct	Example	What full support adds
selectordinal	{n, selectordinal, one {#st} two {#nd} few {#rd} other {#th}}	ordinal plural categories (1st/2nd/3rd), distinct from plural's cardinal set; delegable to Intl.PluralRules(tag, {type: "ordinal"})
plural offset:	{n, plural, offset:1 one {you and # other} other {you and # others}}	subtract an offset before category selection and # formatting
plural exact-value arms	{n, plural, =0 {none} one {# item} other {# items}}	literal =N matches that bypass category selection
CLDR skeleton strings	{n, number, ::currency/GBP} / {d, date, ::yMMMd}	arbitrary Intl options beyond the fixed style keywords (integer/percent; short/medium/long/full) — including currency, which needs a currency code the placeholder can't carry today
nested {arg, …} dispatch	{n, plural, other {{g, select, …}}}	a second format dispatch inside a sub-message; today a sub-message is literal text + # only, one dispatch level per placeholder

Notes for whoever picks one up

• selectordinal, the offset:/=N variants, and the fixed→skeleton widening are all host-Intl-delegable (ADR 0276 Decision A) — no CLDR data in the compiler. The skeleton case is the largest: it re-opens the "how much Intl option surface to expose" question and, for currency, needs a value/type to carry the currency code (couples to the Money kernel, which has no formatting method — ADR 0276 context).
• Nested dispatch is the one real parser change (icu.rs currently caps sub-messages at literal + #); the others extend the existing plural/number/date arms.
• Each currently rejects via bynk.messages.malformed_icu_syntax — a proposal that builds one narrows that code's scope; add a positive fixture and (if Intl-delegated) a workerd smoke test, mirroring slice 3.

Done when

Each construct, when demanded, is filed as its own proposal citing this issue; there is no work to do on this issue itself beyond closing it if the exclusions are ever all lifted.