feat(search): let a search type name its label field

[ddeboer]

A type serving as a labelSource had to declare its display field with the
literal name label, so the word the API serves was decided by an internal role
rather than by the profile the schema declares to. In SCHEMA-AP-NDE, Person,
Organization, Place and DefinedTerm all carry schema:name, and consumers
expect name – but only a type nothing resolves labels from could serve it.

A type can now name its own label field:

export const term = defineSearchType({
  name: 'Term',
  class: `${SCHEMA}DefinedTerm`,
  labelField: 'name',
  fields: [{ name: 'name', kind: 'text', /* … */ }],
});

Changes

• @lde/search: SearchTypeBase gains labelField; labelFieldOf resolves it
  through the new labelFieldNameOf (with DEFAULT_LABEL_FIELD), and the
  schema-wide validation message quotes the configured name. Both helpers are
  exported from the adapter surface.
• @lde/search-api-graphql: a reference object type keys its resolved label off
  the label source’s configured name, so TermReference becomes { id, name }.
  ValueBucket.label is unchanged – it is per-facet-field, and a per-type name
  would make the bucket type non-uniform.
• Since one emitted reference type cannot serve two words, buildGraphQLSchema
  now rejects fields that share a ref.typeName but resolve labels from sources
  disagreeing on their labelField – otherwise declaration order would silently
  decide which word both fields serve.
• @lde/search-typesense needs no change: it already resolves the field through
  labelFieldOf. The engine-port property label also stays – it is a wire
  name, not a served one.
• Docs: a ‘Naming the label field’ section in the search reference, plus updates
  to the GraphQL and Typesense references, packages/search/CONTEXT.md, and
  amendments to ADR 4 and ADR 8.

Non-breaking – every existing schema keeps the default.

Also fixes an unrelated test that this branch happens to make CI run: the
Prettier-peer case in print-sdl.test.ts mocked a module the earlier tests had
already loaded, so it passed locally and failed on CI. It now resets the module
graph and re-imports the module under test.

Fix #727