fix(opencode): preserve reasoning_content for Moonshot/Kimi tool-call assistant messages

[XenoAmess]

Issue for this PR

Closes #29619

Type of change

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

What does this PR do?

to fix this:

thinking is enabled but reasoning_content is missing in assistant tool call message at index xxx

see this

BerriAI/litellm#21672

How did you verify your code works?

we re-ask the failed question and this time it can work.

Screenshots / recordings

after the fix it can work.

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 your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential Related PR Found:

• PR fix(llm): forward reasoning_content in experimental OpenAI Chat assistant messages #28346: fix(llm): forward reasoning_content in experimental OpenAI Chat assistant messages
  • URL: fix(llm): forward reasoning_content in experimental OpenAI Chat assistant messages #28346
  • Relation: This PR addresses a similar issue of forwarding reasoning_content in assistant messages but for OpenAI Chat. The current PR (31995) handles the same concern for Moonshot/Kimi providers. These are likely addressing the same underlying issue across different LLM providers.

Additionally, there are related PRs that deal with reasoning/thinking content preservation:

• PR fix: preserve thinking/redacted_thinking blocks in Anthropic message transforms #23755: fix: preserve thinking/redacted_thinking blocks in Anthropic message transforms
• PR fix(provider): skip empty reasoning_content to preserve KV cache hits #28352: fix(provider): skip empty reasoning_content to preserve KV cache hits

These suggest an ongoing effort to properly handle reasoning content across various LLM providers.

[github-actions]

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