Function load_blocking 

pub fn load_blocking(
    worker: &GpuWorker,
    model_name: &str,
    config: &Config,
) -> Result<()>

Synchronously load a model on this GPU worker for the admin API.

Acquires the per-GPU load lock, then delegates to ensure_model_ready_sync. Intended to be called inside tokio::task::spawn_blocking. Uses the size-only peak (no resolution context) for the preflight — admin loads don’t carry a request shape.