Narrow rune schemas to applicable universal attributes

Approach

This is the breaking change. {% card reading="prose" %} moves from a silent no-op to a Markdoc validation error. Intended — the author gets no feedback at all today — but it will fail builds on existing content.

The migration is decided: narrow directly

SPEC-125 deferred the choice between narrowing outright and a transitional annotate-only minor until "Phase 1 has quantified the real-world impact". It has, and the answer is that there is almost none: a scan of all 986 markdown files in the repo found 25 uses of a gated universal attribute in live content, none of which a narrowed schema would reject. The five that would be rejected are all inside documentation code fences, and were corrected in v0.31.0 rather than migrated.

So the transitional minor buys nothing and costs a release of the silent no-op it is meant to replace. Narrow here. The spec's Migration section carries the numbers and the two doc corrections the scan surfaced.

Note that this measures this repo. Downstream sites may hold content the scan cannot see, which is what the breaking changeset is for — but the shape of the break is now known to be narrow, not broad.

declaredSlots() is not a gating input

WORK-532 left a shared declaredSlots(config) helper and predicted this item would want it. Use it for diagnostics if useful, but not to decide applicability: it reads slots out of layout/structure/autoLabel/blocks/ contentWrapper and is deliberately a lower bound, because slots a schema emits as bare data-name attributes are invisible from config. Card's title is one such slot. Gating comes from sections and modifiers, which are now reachable from the schema via schemaRuneStructures (WORK-533).

Exposure is probably low: most of it is content setting an attribute that never did anything, on a rune that never could honour it, so the likely source is copy-paste between runes. And v0.31.0 removed the sharpest cases in advance — the runes an author is most likely to have written reading on are the six whose roles were missing, and those now work rather than erroring.

Before committing to rejection, measure. ADR-028 keeps a transitional alternative open: keep schemas permissive for one minor and have the tooling annotate rather than reject, narrowing in the release after. Decide from the real numbers, not from the estimate above.

The "declared slots for a rune" helper from WORK-532 is likely reusable here.