feat(provider): add Eden AI as a first-class provider

[Nosenzor]

Issue for this PR

Closes #

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds Eden AI (https://www.edenai.co) as a built-in provider, wired the same way as OpenRouter so it works with no manual provider block in opencode.json. Eden AI is an OpenAI-compatible gateway (POST /v3/chat/completions, Authorization: Bearer, EDENAI_API_KEY) that routes to many models behind one key, including the @edenai smart router.

The catch: the provider catalog is sourced entirely from models.dev, and Eden AI isn't published there yet. Instead of blocking on an upstream models.dev entry, Eden AI ships as a models.dev-shaped record in packages/core/src/models-dev-builtin.ts and is merged into the catalog via a small withBuiltinProviders() helper at each place the catalog is assembled — the v1 Provider service, the v2 Catalog, the connect/list HTTP handler, and the CLI listing commands. I deliberately kept ModelsDev.get() itself untouched so it stays a faithful mirror of models.dev (its unit tests pin that contract), which is why the merge lives at the consumers rather than in the service. The merge is { ...builtin, ...data }, so if Eden AI ever lands on models.dev the real entry wins and the built-in silently steps aside.

The rest mirrors how OpenRouter is wired:

• identity headers (HTTP-Referer / X-Title) in both the legacy CUSTOMIZE map and a new v2 EdenAiPlugin;
• an edenai profile + provider in packages/llm (OpenAI-compatible chat, with fallbacks / routerCandidates passthrough);
• UI wiring: provider icon, popular-providers entry, and a connect note.

Like OpenRouter it's connect-gated — it only shows up once EDENAI_API_KEY is present. Default models are the @edenai smart router plus a few popular provider/model ids (Claude, GPT-5, Gemini); their costs/limits are indicative list prices, since Eden AI reports exact per-request cost at runtime.

How did you verify your code works?

• typecheck passes across core, opencode, llm, schema, app, ui, and server.
• Tests: the full core suite and the legacy opencode provider/config/server suites pass. Added unit tests for the llm provider (packages/llm/test/provider/edenai.test.ts) and the v2 plugin (packages/core/test/plugin/provider-edenai.test.ts), and updated the models-dev and exports tests for the built-in.
• Ran the CLI against the built-in in isolation (empty config, OPENCODE_DISABLE_MODELS_FETCH=true, EDENAI_API_KEY set): opencode models edenai lists exactly the six built-in models, and with no key it's hidden — identical to openrouter.

Screenshots / recordings

The only UI surface is the Eden AI entry in the connect/model dialogs (icon + name + note); there are no layout changes.

Providers → connected, showing Eden AI with its icon:

Popular providers, Eden AI listed with its connect note ("Route to 500+ models with smart routing and fallbacks"):

Connect dialog, prompting for EDENAI_API_KEY:

Models, the Eden AI section listing the routed models (Smart Router + provider/model ids):

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR