LLM: Switch default to OpenAI + healthcheck + streaming endpoint

[BossyT]

Goals

• Provide a thin OpenAI client that reads configuration from environment variables.
• Expose a streaming chat endpoint at /api/ai/chat with raw text output using Edge runtime.
• Expose a health endpoint at /api/health reporting status of the OpenAI service.
• Update documentation with required environment variables and show example usage.

Acceptance Criteria

• A new file lib/ai/openai.ts implements streamChat() using OPENAI_API_KEY, OPENAI_MODEL, OPENAI_BASE_URL and AI_PROVIDER=openai environment variables.
• A new API route /api/ai/chat in the Edge runtime accepts chat messages { messages, model?, temperature? } and streams back plain text output. Missing AI_PROVIDER=openai returns an error.
• A new API route /api/health returns JSON { status, services: { openai: { ok, model, error? }}}; it uses OPENAI_API_KEY to validate connectivity with OpenAI, and sets ok: false with a helpful error when the key is missing or invalid.
• The README explains required environment variables (AI_PROVIDER, OPENAI_API_KEY, OPENAI_MODEL, OPENAI_BASE_URL) and includes simple instructions for streaming chat and health check.
• The feature does not crash the app when secrets are missing; openai.ok is set to false with an error in /api/health.
• Include screenshots of the new /api/health endpoint (showing both successful and missing-key responses) and a sample streamed chat in the PR description.