0.5.0

Added

• Settings are grouped into collapsible sections (Missions / Appearance / CIPHER uplink);
  the open/closed state is remembered.
• The CIPHER endpoint is now an ordered fallback list instead of a single URL — the first
  reachable one is used. One synced list covers the same local LLM server showing up as
  localhost at your desk and as a LAN IP on the road. Existing single-endpoint settings
  from 0.4.x migrate automatically.
• New "Model thinking" toggle (off by default): CIPHER answers straight away instead of
  deliberating. Models that always think (gpt-oss/harmony) are detected and the toggle
  disables itself instead of promising something the request can't deliver.
• The selected model's context length is shown when the endpoint reports it (LM Studio,
  Ollama).