Update extensions and documentation for LLM integration and CLI enhancements

- Added support for a new extension, `swarm-assistent`, in `extensions.example.yaml` with a requirement for `ollama`.
- Enhanced the README.md to clarify the setup process and provide a quick start guide for using extensions.
- Updated documentation in `llm.md` to reflect the opt-in nature of LLM support and provide clearer instructions for enabling it.
- Improved the `autocomplete.md` to detail the automatic setup of word lists during the initial launch.
- Revised `cli.md` to include new commands and options related to LLM runtime handling and extension management.
- Enhanced the `spike-notes.md` to guide users through the first live run with a focus on LLM integration.

docs/architecture.md

@@ -17,8 +17,9 @@
 ┌─ Selectel, сегмент пула (например ru-7a) ────────────────────┐
 │  GPU VM (tag preemptible + gpu-rent)                          │
 │    SwarmUI : 127.0.0.1:7801                                   │
-│    idle-killer: очередь / hold / качалка в UI → delete self   │
-│    application credential (compute delete/shelve only)        │
+│    idle-killer: очередь / hold / LLM busy → delete this server │
+│    app cred: DELETE/GET только этот server_id (fail closed)   │
+│    optional: Ollama :11434 / llama.cpp :8080 (туннель 17811/12)│
 │                                                               │
 │  boot volume (network)     ОС + NVIDIA + SwarmUI нативно + snapshot │
 │  data volume (network)     Models, Output, Data, workflows     │