Function worker_loop

One worker thread: owns a CUDA context on `gpu` and a single pair of reusable in/out buffers. Loops, pulling seeds from `stats.next_seed`, until either `max_iters` or `stats.stop_at_iter` is hit (we hit the stop sentinel by setting it to 0 on Ctrl-C — see main).

(
    worker_id: usize,
    gpu: i32,
    cfg: &Config,
    arch: &str,
    stats: &Stats,
    log_lock: &Mutex<()>,
)