Add Gemini-to-OpenAI Chat compatibility helpers, tests, and CI job

[hungvminh]

Motivation

• Provide a small compatibility layer to translate Gemini-style generate responses to OpenAI Chat Completions shapes for easier integration.
• Make the conversion deterministic for downstream code by normalizing message roles, parts, and finish reasons.
• Run and enforce the compatibility tests in CI to catch regressions early.

Description

• Added a new helper module at server/src/modules/common/geminiChatCompat.ts that exposes toGeminiContents(messages) and toOpenAIChatCompletion(geminiResponse, model, id?) and relevant types.
• Implemented finish-reason normalization mapping Gemini finish reasons (STOP, MAX_TOKENS, SAFETY) to OpenAI-style reasons (stop, length, content_filter).
• Added unit-style assertions in server/src/modules/common/geminiChatCompat.test.ts that validate conversions and usage/choice shaping.
• Updated server/package.json with a test:gemini-compat script and documented the helper in README.md, and added a GitHub Actions workflow .github/workflows/ci.yml to run the test under node 20.

Testing

• Ran the new automated test via npm run test:gemini-compat in the server package and the assertions completed successfully.
• The CI workflow gemini-compat-tests was added to execute the same npm run test:gemini-compat step on ubuntu-latest with Node.js 20.
• The test file logs geminiChatCompat tests passed on success and uses built-in assert checks to validate conversions.

---

Codex Task