New 'llm' bot type that takes a startup context (system prompt) and replies to
each message via an OpenAI-compatible endpoint — works with a local Ollama
(ollama serve, http://localhost:11434/v1), OpenAI, Grok, etc. Generalize the
support LLM handler into _handle_llm_message (shared by support + llm) with a
per-bot default prompt. Create form reuses the LLM fields (URL/key/model/context)
for both support and llm. Adds llm_test.py (mock OpenAI backend) — passes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
