fix(provider): avoid incompatible GPT-5 defaults and sanitize Gemini tool schemas

[Haohao-end]

Issue for this PR

Closes #18882

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

This fixes two compatibility problems from #18882.

ProviderTransform.options() was applying GPT-5 default request fields too broadly based on the model ID. That caused generic @ai-sdk/openai-compatible providers or proxies using models like gpt-5.4 to receive fields such as reasoningSummary, reasoningEffort, textVerbosity, and include, even when the backend did not support them. This change limits those defaults to transports/providers that explicitly support them.

This PR also tightens Gemini tool schema sanitization. The Gemini schema path now inlines local refs and recursively removes unsupported keys such as $schema, additionalProperties, $ref, ref, $defs, and definitions. Structured output now reuses the same sanitization path so it does not bypass the Gemini-specific cleanup.

How did you verify your code works?

I added regression tests covering:

• generic @ai-sdk/openai-compatible gpt-5.4 models no longer receiving OpenAI-only defaults
• opencode* GPT-5 providers still keeping supported defaults
• recursive Gemini schema sanitization, including local ref inlining
• structured output using the same Gemini schema sanitization path

I also ran:

• git diff --check
• cd packages/opencode && bun run typecheck
• bunx turbo typecheck

Screenshots / recordings

Not a UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

If you do not follow this template your PR will be automatically rejected.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍