- Add set_parallel(n) to LlamaCppBackend: kills and relaunches llama-server
with updated --parallel flag (stores launch cmd/env for replay)
- Add POST /access-endpoint/enable: restarts with --parallel 2, generates key
- Add POST /access-endpoint/disable: restarts with --parallel 1, clears key
- GET /access-endpoint now returns enabled state
- Model load clears previous API key (user must explicitly enable)
- Frontend dialog shows enable/disable toggle with loading state
- Connection details (URLs, key, snippets) only shown when enabled