feat: add Ollama LLM provider - #7

Fixes #2.

Adds an OllamaClient that talks to the local OpenAI-compatible /v1/chat/completions endpoint, plus make_client() routing for ollama/<model> so users can run PromptQuery without an external API key or schema data leaving their machine.

The implementation deliberately uses Python's stdlib HTTP/JSON stack instead of adding a runtime dependency. OLLAMA_BASE_URL can override the default http://localhost:11434 endpoint.

Validation

• .venv/bin/pytest tests/test_llm.py -q -> 5 passed
• .venv/bin/pytest -q -> 54 passed
• git diff --check -> clean