A pinned model no longer splits its provider's group in the model picker

The model picker groups models by provider. Promoting one model within a provider — so it becomes the deployment's default — could push that single model out of its provider's run in the list, making the provider render as two separate headers with the promoted model stranded between them. Provider is now the primary grouping key, so every provider's models always render as one contiguous block; a promotion still moves the model to the top of its own group, exactly as intended, without tearing the group apart.