Delivery to OpenAPI, MCP and LLM tooling (x-jsonld-* derivation)

[simontaurus]

Completes #42 (gap analysis #18, finding C3; prefixing already landed in #44): the @context -> REST-API-LD x-jsonld-* derivation and the tiered delivery model.

New "Delivery to OpenAPI, MCP and LLM tooling" subsection under Usecases.

• Goal: keep the schema's semantics available to the consumer (grounding + downstream RDF), not to emit @context in output. With additionalProperties: false, @context stays schema-level metadata.
• MCP / tool-use / structured output: a top-level @context is carried through and used as grounding - no relocation needed. Confirmed by a real MCP JSON-RPC roundtrip: a server advertised a tool whose inputSchema had two opaque properties a/b mapped via @context to schema:familyName/schema:givenName; the client received the @context intact over tools/list, and the model filled {a: "Mustermann", b: "Max"} (by IRI, not surface order). MCP's inputSchema/outputSchema (2025-06-18) impose no keyword restrictions.
• OpenAPI: 3.1 takes @context as-is; 3.0 forbids unprefixed keywords and, when bundling several classes under components/schemas, has no document root for one @context. Both solved by REST-API-LD's per-Schema-Object x-jsonld-context / x-jsonld-type. OO-LD maps @context -> x-jsonld-context and x-oold-instance-rdf-type -> x-jsonld-type per class.
• Strict subsets that reject unknown keywords: fold IRIs into title/description.

Summary table + TOC entry added. No example changes; 5/5 pass CI.