session_setup: show resolved default model in chat status line

[TroyHernandez]

Summary

corteza::chat() displayed `(provider default)` as a placeholder in the status line when no model was specified. Now resolves the provider's default upfront so callers see the actual model name.

Before / After

```
corteza chat | (provider default) @ moonshot | 30 tools | /clear, /quit
```

becomes

```
corteza chat | kimi-k2 @ moonshot | 30 tools | /clear, /quit
```

Change

One line in `R/session_setup.R` adds a third fallback after the user-passed model and the project config's model:

```r
model <- model %||% config$model %||% llm.api::provider_default_model(provider)
```

Uses `llm.api::provider_default_model()` (added in llm.api 0.1.2.1, cornball-ai/llm.api#10) so corteza doesn't reach into llm.api internals.

The defensive `%||% "(provider default)"` placeholders in `R/chat.R:240` and `:340` are left in place — they're now unreachable under normal use, but a useful safety net if some future channel bypasses `session_setup`.

Test plan

• All 1235 corteza tests pass locally
• Verified end-to-end: `session_setup(cwd = "/home/troy/tiny")` resolves `model_map$cloud` to `"kimi-k2"` with the moonshot config

[TroyHernandez]

Blocked on llm.api 0.1.3 reaching CRAN. provider_default_model() was added on llm.api main but isn't on CRAN/r2u yet, so CI installs the older 0.1.1 and fails. Holding this draft until the dependency lands.