Accelerate DeepSeek DSpark inference

src/engine/metal/gpu.rs

@@ -1644,6 +1644,13 @@ impl Commands {
             "executing Metal commands",
         )
     }
+
+    pub(super) fn flush(&mut self) -> Result<(), String> {
+        check(
+            unsafe { ds4_gpu_flush_commands() },
+            "flushing Metal commands",
+        )
+    }
 }
 
 pub(super) struct ParallelFfn(bool);