AIServicePlugin should auto-detect and initialize LLM provider based on environment variables

[hotlong]

Problem

Currently, only the CLI server (packages/cli/src/commands/serve.ts) contains logic to auto-detect LLM providers (OpenAI, Anthropic, Google, Vercel AI Gateway) via environment variables and construct a VercelLLMAdapter accordingly. In non-CLI deployments (e.g., Vercel via apps/studio/server/index.ts), AIServicePlugin is always constructed with no adapter supplied, so it falls back to the test-only MemoryLLMAdapter.

This leads to inconsistent behavior: production deployments risk silently using only the echo AI, unless every entrypoint duplicates the CLI detection logic.

Proposal

Refactor and move the entire LLM provider auto-detection and adapter selection logic into the core AIServicePlugin. When no adapter is supplied in options, AIServicePlugin should:

1. Detect LLM provider preference and settings from standard environment variables, in priority order:
   • AI_GATEWAY_MODEL → use @ai-sdk/gateway via VercelLLMAdapter
   • OPENAI_API_KEY → use @ai-sdk/openai via VercelLLMAdapter
   • ANTHROPIC_API_KEY → use @ai-sdk/anthropic via VercelLLMAdapter
   • GOOGLE_GENERATIVE_AI_API_KEY → use @ai-sdk/google via VercelLLMAdapter
2. Fallback to the MemoryLLMAdapter only if no config is available
3. Log which provider is selected or a warning if falling back to memory adapter

Accept dynamic import failures as a soft error: optional peer dependencies should not break boot, just emit a warning and try the next provider.

Benefits

• No need to repeat auto-detect logic for every serverless or containerized entrypoint
• All use cases (CLI, Vercel, Docker, custom node) consistently get best-available real AI
• One place to extend with new providers in the future
• Reduced operational risk from silent fallback

Tasks

• Refactor LLM adapter auto-detection into AIServicePlugin (private method during init)
• Remove redundant LLM detection logic from CLI serve.ts
• Add logging to report which adapter/provider is active
• Add unit test (mock env) for provider detection logic
• Update roadmap and changelog