Validate every response against the OGC schemas

[estebanzimanyi]

Annex A's prose says what a response must carry; the schemas say it in a form a machine checks. A suite that asserts the fields it thought to name passes a document missing everything it forgot, so the schema is what makes the conformance claim mean something.

openapi/ogcapi-movingfeatures-1.bundled.json is the OpenAPI document OGC publishes for Part 1, vendored byte for byte and never edited. It is the bundled form, so every $ref resolves inside the one file and validation needs no network — it runs in the same job as every other test. A test guarded by MFAPI_SCHEMA_FRESHNESS re-fetches the URL and compares. The landing page, the conformance declaration, a Collections document, a Collection, a TemporalProperties document and a TemporalGeometrySequence are each validated against the schema the standard names for it.

One translation is applied and its count asserted: OpenAPI 3.0 spells nullability as nullable: true beside a type where a JSON Schema draft spells a type union, and a validator that does not know the keyword would quietly ignore it. Asserting the count keeps the rewrite from becoming a no-op, since a translation that stops applying makes the run more permissive without saying so. Nothing else is rewritten — measured on this document, no schema puts a sibling keyword beside a $ref, and neither exclusive bound appears in the boolean form OpenAPI uses.

format is asserted rather than annotated, and TestATSSchemaMotionCurveIsInverted measures why. At JSON Schema's default reading, the standard's own motionCurve rejects all five interpolation values it names and accepts an arbitrary string, because its two oneOf branches both match a plain string:

value	published schema, default format	format asserted
Discrete … Cubic	rejected	accepted
http://example.org/curve	accepted	accepted
Bogus	accepted	rejected

The same test measures the correction proposed for it — an absolute-URI pattern on the URI branch — which answers correctly under both readings, where anyOf continues to accept Bogus.

Two further tests record what the standard says about itself rather than asserting a constant somebody chose: the temporalGeometrySequence enum that settles Annex A's contradiction, and the fact that a temporal geometry's datetimes carry no format where a temporal property's carry date-time. The second is a gap in this validation rather than a finding about the tier, and it is stated so a passing TemporalGeometrySequence is not mistaken for a checked one.