fix: fall back to target provider's default model on provider switch

[cursor]

What Changed

When switching providers (e.g., Codex to Claude) without explicitly specifying a model, the ensureSessionForThread function now falls back to DEFAULT_MODEL_BY_PROVIDER[targetProvider] instead of carrying over the thread's existing model slug.

The unit test expectation is updated accordingly to assert "claude-sonnet-4-6" instead of "gpt-5-codex" when provider: "claudeAgent" is requested.

Why

desiredModel was computed as options?.model ?? thread.model. When a user switches from Codex to Claude without providing a model override, the thread's stored model (a Codex-specific slug like "gpt-5-codex") was forwarded to the Claude adapter unchanged. The Claude SDK doesn't understand Codex model names and will fail or behave unpredictably.

The fix detects when options.provider is explicitly set and differs from the current provider (including when there is no current session), and uses the target provider's default model from the existing DEFAULT_MODEL_BY_PROVIDER map.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes — N/A, no UI changes
• I included a video for animation/interaction changes — N/A

  

Note

Fix ProviderCommandReactor to use target provider's default model on provider switch

When switching providers without specifying an explicit model, the reactor previously kept the thread's existing model (from the old provider). It now selects the default model for the incoming provider via DEFAULT_MODEL_BY_PROVIDER.

The fix adds a providerIsChanging check in ProviderCommandReactor.ts to distinguish between a same-provider resume and a cross-provider switch when resolving desiredModel.

^{Macroscope summarized 5b9f3fb.}