[Feature] Add Vercel AI Gateway as a built-in provider

I work on Vercel's AI Gateway. We verified ZCode 3.9.1 works with it end to end as a custom provider, and we're adding ZCode to our vercel ai-gateway coding-agents setup CLI command — a built-in entry on your side would complete the loop.

Verified on 3.9.1 (macOS arm64), all with real agent tasks including tool calls:

• Chat completions (/chat/completions): Base URL https://ai-gateway.vercel.sh/coding-agent/v1 — works, including the reasoning effort levels
• Responses (/responses): Base URL https://ai-gateway.vercel.sh/codex/v1 — works
• Anthropic messages (/v1/messages): Base URL https://ai-gateway.vercel.sh — endpoint and auth work, but Claude models are blocked by the unconditional thinking.type: "enabled" injection tracked in #233

One key gives access to 147 models across providers with unified billing. ZCode already runs on Vercel's AI SDK (ai-sdk/provider-utils is in the request user-agent), so the wiring should be familiar: the chat-completions config above, plus model discovery from GET /coding-agent/v1/models — the catalog returns context_window, max_tokens, and modalities per model, which maps directly onto your model config fields.

Happy to test builds or send anything else you need.

Start by locating the existing custom-provider wiring and the ai-sdk/provider-utils request path. Use the supplied chat-completions base URL and GET /coding-agent/v1/models response as the integration reference; done means the gateway is available as a built-in provider with model metadata mapped into the existing config fields.