- GET /models/available: query both llama-swap instances for model lists
- POST /models/select: set per-persona model (regular/evil/japanese) with persistence
- GET /models/status: return current per-persona model assignments
- Fall back to known model list when containers are unreachable
