Unified provider sends `max_tokens` to OpenAI reasoning models

`ai-gateway-provider`’s Unified route currently creates its models via `@ai-sdk/openai-compatible`:

```ts
const unified = createUnified();
const model = aigateway(unified('openai/o1-mini'));
```

When used with AI SDK’s standardized `maxOutputTokens`, the outgoing Chat Completions request contains:

```json
{
"model": "openai/o1-mini",
"max_tokens": 1000
}
```

OpenAI reasoning models reject this field and require `max_completion_tokens` instead:

```text
Unsupported parameter: 'max_tokens' is not supported with this model.
Use 'max_completion_tokens' instead.
```

The native Cloudflare OpenAI adapter does not have the issue:

```ts
const openai = createOpenAI();
const model = aigateway(openai.chat('o1-mini'));
```

That route uses `@ai-sdk/openai`, which performs the correct OpenAI-specific parameter mapping.

Suggested fix: have `createUnified()` supply `transformRequestBody` to its `createOpenAICompatible` instance. When Cloudflare’s model-capability metadata identifies an OpenAI reasoning model, move `max_tokens` to `max_completion_tokens`.

Commencez par createUnified() et son instance createOpenAICompatible, puis suivez la manière dont transformRequestBody reçoit les métadonnées de capacité du modèle de Cloudflare et mappe maxOutputTokens standardisé. Reproduisez la requête openai/o1-mini et vérifiez que les modèles de raisonnement envoient max_completion_tokens, tandis que les autres modèles conservent leur comportement actuel des paramètres.

Rédigé par le modèle d'indexation à partir du texte de l'issue.