feat(openrouter): model fallbacks and routed-model exposure

## Summary

Add typed OpenRouter model fallback routing via a `ModelFallbacks` helper and expose the provider-reported routed model through completion, streaming, and agent response surfaces.

## Motivation

OpenRouter supports a `models` fallback array for chat completions. Rig currently has no typed API for this (unlike `ProviderPreferences`), and the response `model` field (which may differ from the requested primary when fallbacks fire) is not surfaced to generic callers or agent `.prompt()` users.

## Proposed scope

- `ModelFallbacks::new([...])`, `to_json()`, and `CompletionModel::with_model_fallbacks()`
- Serialize `{ "model": "", "models": ["", ...] }` with documented precedence rules
- Populate `CompletionResponse.response_model` and propagate to agent `CompletionCall` / `PromptResponse` / streaming `FinalResponse`
- Unit tests for serialization/precedence; ignored live smoke tests for OpenRouter

Embeddings/transcription fallbacks are out of scope.

Look for the OpenRouter provider implementation and the CompletionModel struct. The work involves adding a ModelFallbacks helper, integrating it with CompletionModel::with_model_fallbacks, and ensuring the response model field is propagated through CompletionResponse and agent surfaces like PromptResponse. Check existing tests for serialization patterns and add unit tests for the new fallback logic. A live smoke test for OpenRouter should be marked as ignored.

Scritto dal modello di indicizzazione a partire dal testo della issue.