fix: handle OpenRouter reasoning details for Gemini models

[TheSethRose]

Description

This PR fixes an issue where Gemini models via OpenRouter would fail with a 400 error when using tools. The error was caused by missing reasoning_details (thought signatures) in subsequent requests, which OpenRouter/Google AI Studio requires to be preserved and replayed.

Error fixed:
Gemini models require OpenRouter reasoning details to be preserved in each request. ... Function call is missing a thought_signature in functionCall parts.

Changes:

• Implemented OpenRouterEndpoint to capture reasoning_details from streaming responses.
• Added a reasoningCache in OpenRouterLMProvider to associate reasoning blocks with tool call IDs.
• Injected cached reasoning_details back into outgoing request messages for assistant turns containing tool calls.
• Added comprehensive unit tests in src/extension/byok/vscode-node/test/openRouterProvider.spec.ts.

Testing

• Added unit tests covering:
  • Model fetching and parsing.
  • Reasoning detail caching and cleanup.
  • Injection of reasoning details into request bodies.
• Verified the fix by compiling the extension and testing the VSIX.

Checklist

• Signed CLA (Verified via GitHub profile)
• Tests added/updated
• Followed project architecture and coding standards
• Verified compilation output

Fixes: Copilot Request id: 8aa4dd19-4578-4e8c-abe8-986364545cd9

[TheSethRose]

@microsoft-github-policy-service agree