feat(provider): add Yutori provider support for Navigator model

[dhruvbatra]

Summary

Adds Yutori as a model provider so openbench can target Yutori's Navigator computer-use model family. The Navigator API is OpenAI-compatible Chat Completions, so the new YutoriAPI plugs into Inspect AI's standard OpenAICompatibleAPI machinery — same pattern as Baseten, SiliconFlow, etc.; closely modeled on #269.

Navigator is a screenshot-driven vision/computer-use model, not a text reasoner. This PR is just the provider plumbing — the natural pairing is a browser-use or vision benchmark (bring your own with bench eval path/to/eval.py, or wait for a future built-in).

What are you adding?

• New model provider

Changes Made

• New YutoriAPI(OpenAICompatibleAPI) in src/openbench/model/_providers/yutori.py.
• @modelapi(name="yutori") registration in _registry.py; ProviderType.YUTORI + entry in PROVIDER_CONFIGS.
• Provider row in README.md; provider row + "Yutori-Specific Configuration Options" subsection in docs/providers.mdx.
• 21 unit tests in tests/test_yutori_provider.py.

Navigator-specific behavior in the provider

• Navigator's Chat Completions API uses max_completion_tokens (the modern OpenAI Chat Completions field) and rejects max_tokens. The provider's completion_params() silently remaps openbench's --max-tokens flag onto max_completion_tokens so callers don't need to know about the rename. An explicit -M max_completion_tokens=N wins over the remapped value.
• A few Navigator-only request fields are exposed as -M kwargs and forwarded as extra_body: tool_set, disable_tools, json_schema (matching the public Yutori SDK surface).

Testing

• pytest tests/test_yutori_provider.py — 21 passed
• pytest — full non-integration suite passing
• pre-commit run --all-files — ruff, mypy, registry-imports all green
• Smoke-tested: inspect_ai.get_model("yutori/n1.5-latest") resolves to YutoriAPI, picks up YUTORI_API_KEY and YUTORI_BASE_URL overrides correctly

Checklist

• My code follows the project's style guidelines
• I have performed a self-review of my own code
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my feature works
• New and existing unit tests pass locally with my changes