Allow LM Studio provider to run in no-key mode for AI smoke and local narration flows - #34

/api/ai/test_provider failed for provider=lmstudio when OPENAI_API_KEY was unset/empty, despite LM Studio being reachable and expected to work in local no-key mode. This blocked setup QA and contradicted local OpenAI-compatible behavior expected for demo/release verification.

• LM Studio-only credential fallback

  • Updated _openai_client_for() to apply a dummy fallback key only for provider == "lmstudio" when OPENAI_API_KEY is empty/missing.
  • Kept existing credential behavior unchanged for all other providers.

• Regression coverage for provider-specific auth behavior

  • Added contract test validating POST /api/ai/test_provider succeeds for LM Studio with empty env key.
  • Added parameterized regression test confirming non-LM Studio providers (openai, ollama, openrouter) retain existing env-key pass-through behavior.

def _openai_client_for(provider: str, base_url: str | None):
    final_base_url = resolve_provider_base_url(provider, base_url)
    if provider and provider.lower() == "lmstudio":
        api_key = os.environ.get("OPENAI_API_KEY") or "lmstudio-local"
    else:
        api_key = os.environ.get("OPENAI_API_KEY", "dev-local")
    return OpenAI(api_key=api_key, base_url=final_base_url)