fix(google_genai): strip id from FunctionCall/FunctionResponse wire format

[n-WN]

Related Issue

N/A — discovered through manual testing with Gemini models via third-party proxy endpoints.

Description

Gemini API returns HTTP 400 when id is included in FunctionCall or FunctionResponse parts of the request body. This causes all Gemini models to crash after the first tool call — the first step succeeds, but the second step (which replays the conversation history containing function_call with id) is rejected by the API.

Root Cause

message_to_google_genai() and _tool_message_to_function_response_part() passed id=... to FunctionCall() / FunctionResponse() constructors. The google-genai SDK accepts the kwarg, but the Gemini REST API rejects it on the wire:

400: Invalid JSON payload received.
  Unknown name "id" at 'contents[1].parts[0].function_call': Cannot find field.
  Unknown name "id" at 'contents[2].parts[0].function_response': Cannot find field.

Gemini matches function calls to responses by array position, not by id (unlike OpenAI). The internal tool_call_id tracking in kosong's Message objects is unaffected — only the outbound wire format changes.

Fix

Remove id=... from FunctionCall and FunctionResponse constructors (2 lines). The SDK's model_dump(exclude_none=True) serialization automatically omits None fields, so the id key no longer appears in the wire JSON.

Manual Verification

Tested on a machine with 6 Gemini model configurations through qianxun proxy:

Model	Provider	Before	After
gemini-3-pro-preview	google_genai	❌ 400	✅
gemini-2.5-pro	google_genai	❌ 400	✅
gemini-2.5-flash	google_genai	❌ 400	✅
gemini-3-flash-preview	google_genai	❌ 400	✅
gemini-3-pro	vertexai	✅ (unaffected)	✅
gemini-3-flash	vertexai	✅ (unaffected)	✅

Checklist

• I have read the CONTRIBUTING document.
• I have linked the related issue, if any.
• I have added tests that prove my fix is effective or that my feature works.
• I have run make gen-changelog to update the changelog.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.