Fix four real-world read/write bugs found validating against the Stri…

[noha]

…pe API spec

Found while driving OpenAPI-Client against the actual Stripe OpenAPI spec and stripe-mock (not just the Petstore test fixture):

• JSONSchemaArray>>read: computed its result but never returned it (no ^), so array-typed values always silently became the schema object itself. The tuple-validation branch also iterated the schema's own #items instead of the actual data collection.
• JSONSchemaObject>>readObject: sent #ifTrue: directly to #additionalProperties, which crashes whenever a schema sets additionalProperties to a nested schema (or leaves it unset) rather than a literal boolean - both are valid per spec. Stripe's metadata schemas do exactly this.
• JSONSchemaAnyObject had no #read:, hit whenever a property with no fixed shape (e.g. an anyOf collapsing to "accept anything") is read back.
• JSONSchemaString/JSONPrimitiveSchema sent #ifTrue: directly to #nullable, which is nil (not a boolean) for any schema object built by NeoJSON's reflective instantiation - i.e. every schema parsed from a document, primed or not. Changed to a proper accessor that treats unset as permissive, since OpenAPI's "nullable" keyword on a property is not currently threaded through nested schema reads at all (properties are always parsed as base JSONSchemaDefinition, which does not carry it) - a separate, larger gap left for follow-up.