ChatOllama built by esperanto has no timeout; ESPERANTO_LLM_TIMEOUT does not apply to transformations or chat

Summary

ESPERANTO_LLM_TIMEOUT (default 60s) only applies to esperanto's own httpx client. When Open Notebook uses a model through model.to_langchain(), which is the path for transformations, insights and chat, OllamaLanguageModel.to_langchain() builds a ChatOllama without a timeout. langchain_ollama then creates an ollama.Client with timeout=None, so a stalled Ollama request waits forever.

Two consequences:

• The FAQ entry suggesting ESPERANTO_LLM_TIMEOUT=180 for slow local models (docs/6-TROUBLESHOOTING/faq.md) has no effect on transformations or chat.
• A hung Ollama call holds a worker slot (and the SSE connection for chat) indefinitely; the retry logic never gets a chance to run.

Surfaced while investigating #1264.

Where

• Upstream: esperanto/providers/llm/ollama.py — to_langchain() (does not pass client_kwargs={"timeout": ...})
• Here: open_notebook/ai/provision.py / open_notebook/domain/credential.py if we want to pass a timeout from the credential config
• Docs: docs/6-TROUBLESHOOTING/faq.md

Open questions

• Fix upstream in esperanto (pass the configured timeout through client_kwargs) or work around here by setting client_kwargs on the returned ChatOllama?
• Should the timeout be a per-credential setting (like num_ctx) or a global env var?

Acceptance criteria

• A transformation against an Ollama endpoint that never responds fails after the configured timeout instead of hanging.
• FAQ documents which setting actually governs the timeout for the LangChain path.