[Fix] Ollama/OpenAI client compatibility gaps #118

• HEAD / returns 405 (only GET / is defined); Ollama answers "Ollama is running" — health checks fail.
• /api/version falls back to 0.0.0 because the project is not installed (no [build-system]); some clients gate on the version.
• /api/tags lists Hugging Face ids; only qwen3, qwen3:0.6b, qwen3:latest, smollm2:135m map from Ollama names (ollama pull llama3.2 → 404).
• Ollama /api/chat//api/generate do not fetch on first use (unlike the app and OpenAI), and the error tells the user to pull an HF id.
• think: false / /no_think does not stop reasoning: enable_thinking is never passed to the chat template; with think unset raw <think> stays in content.
• README/guide never say how to point Ollama/OpenAI clients at Gemstone (port 23100 vs 11434, OLLAMA_HOST, OPENAI_BASE_URL).

Done when: tests for HEAD / 200, a real /api/version, enable_thinking passed for think: false, first-use fetch (or a consistent policy) on Ollama routes; docs show client configuration.

Found in the 2026-10-03 usability audit (code reading, not run).