Fix Gemini generation config request shape

[g8rr5dg2p7-svg]

/claim #279

Summary:

• Fixes the existing Gemini/PaLM-compatible client so generation settings are sent inside the Google generationConfig request body.
• Exports typed Gemini request/response interfaces from the AI package entrypoint.
• Adds a testcases/palm2 Jsonnet config.
• Adds mocked Vitest coverage for Gemini request shape and chat-style contents.
• Adds a small Jsonnet-driven palm2-chat example.

Verification:

• npm run build
• npx vitest run src/ai/src/tests/palm2/gemini.test.ts
• npx prettier --check src/ai/src/index.ts src/ai/src/lib/gemini/gemini.ts src/ai/src/tests/palm2/gemini.test.ts ../examples/palm2-chat/package.json ../examples/palm2-chat/readme.md ../examples/palm2-chat/src/index.ts ../examples/palm2-chat/src/lib/generateResponse.cts ../examples/palm2-chat/tsconfig.json
• npx tsc --noEmit in JS/edgechains/examples/palm2-chat after package build

Notes:

• Keeps the implementation in the existing GeminiAI class while making the request body match the Generative Language API.
• Prompts for the example live in Jsonnet rather than being hardcoded in TypeScript.