[Feature] Add GPT-6 Astra to OpenRouter built-in support

[Xunzhuo]

Maintainer direction

I recommend accepting this as a bounded Data Plane & Networking feature. #3561 is the reference for a complete model addition, while OpenRouter remains a separate gateway contract whose model identifier, available protocols, reasoning transport, pricing, and conformance must be verified independently.

Goal

Expose openai/gpt-6-astra through the built-in openrouter provider without assuming that every capability of the direct OpenAI endpoint is available through the gateway.

Current-main gap

• config/catalog/resources/providers/openrouter.yaml supports Chat Completions and Responses and declares reasoning_transport: reasoning_object, but it has no GPT-6 Astra model binding.
• OpenRouter currently publishes the gateway model as openai/gpt-6-astra: https://openrouter.ai/openai/gpt-6-astra
• [Feature] Add GPT-6 Astra built-in support #3561 verifies the direct OpenAI wire contract, not OpenRouter's final provider-boundary request shape.

Scope

• Add the openai/gpt-6-astra gateway mapping with the exact current OpenRouter model ID, lifecycle, protocol list, pricing/cache fields, and dated source evidence.
• Verify each advertised protocol and reasoning effort against OpenRouter's current behavior. Do not copy direct-OpenAI protocol support or its top-level reasoning field when the gateway uses a different transport.
• Add final-hop fixtures for every advertised protocol. At minimum, assert model ID, the provider-native reasoning object, absence of competing reasoning fields, and preservation of tool definitions.
• Ensure catalog materialization rejects an effort or protocol that is not supported by this particular binding even when the provider supports it for other models.
• Regenerate the Router, CLI, Dashboard, and website catalog projections and update the Day-0 guide only if the gateway workflow exposes a reusable missing step.

Completion criteria

• The built-in Model Hub shows OpenRouter as a verified GPT-6 Astra provider with current, attributable metadata.
• A catalog-backed OpenRouter configuration materializes the correct upstream model ID without a hand-written model card.
• Provider-boundary conformance proves the exact reasoning and tool request shape for every advertised protocol.
• Invalid protocol/effort combinations fail before provider dispatch.
• Catalog validation, materialization tests, generated-file checks, E2E/provider fixtures, and public documentation pass their owning gates.

Non-goals

• Build a general OpenRouter catalog synchronization service.
• Add unverified protocols, reasoning efforts, or pricing inferred from the direct OpenAI endpoint.
• Change the direct OpenAI implementation in [Feature] Add GPT-6 Astra built-in support #3561.

Reference implementation: #3561.

[Xunzhuo]

/accept