fix: preserve extra_content (thought_signature) for Gemini multi-turn

[gut-puncture]

Fix: Preserve extra_content for Gemini Multi-Turn Tool Calls

Fixes #7519

Problem

When using Gemini 3 Pro via the OpenAI compatibility layer, multi-turn tool call conversations fail with:

400 Bad Request: function call is missing a `thought_signature`

Root Cause

Gemini models include a thought_signature within extra_content.google.thought_signature in tool call responses. This signature must be echoed back in subsequent requests for the model to maintain context. The CLI was discarding this field.

Solution

Added support for preserving and echoing extra_content in the Chat Completions API path:

1. protocol/src/models.rs

Added extra_content field to ResponseItem::FunctionCall:

FunctionCall {
    // ... existing fields ...
    #[serde(default, skip_serializing_if = "Option::is_none")]
    extra_content: Option<serde_json::Value>,
}

2. codex-api/src/sse/chat.rs

Capture extra_content from SSE tool call deltas:

struct ToolCallState {
    name: Option<String>,
    arguments: String,
    extra_content: Option<serde_json::Value>,  // NEW
}

// In processing loop:
if let Some(extra) = tool_call.get("extra_content") {
    call_state.extra_content = Some(extra.clone());
}

3. codex-api/src/requests/chat.rs

Include extra_content in outgoing request JSON:

if let Some(extra) = extra_content {
    tool_call_obj["extra_content"] = extra.clone();
}

Backward Compatibility

• #[serde(default)] - Existing data without extra_content deserializes correctly
• skip_serializing_if = "Option::is_none" - Output stays clean for non-Gemini providers
• Rust .. pattern matching - 20+ existing code paths automatically ignore the new field

Testing

cargo test --package codex-protocol
cargo test --package codex-api
cargo test --package codex-core

Related

• Gemini OpenAI Compatibility: https://ai.google.dev/gemini-api/docs/openai
• Similar fix in openai-agents-js: preserves provider_metadata containing thought_signature

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[gut-puncture]

I have read the CLA Document and I hereby sign the CLA

[gut-puncture]

recheck

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[gut-puncture]

@etraut-openai sir, can we go ahead with the review of this PR or is there any issue here?

[etraut-openai]

@gut-puncture, the team hasn't had time to review this PR yet. There are others ahead of it in the queue.