[Kernel] Update cutlass_scaled_mm to support 2d group (blockwise) scaling (#11868)

csrc/quantization/machete/machete_mainloop.cuh

@@ -272,6 +272,10 @@ struct MacheteCollectiveMma {
   using PipelineState = cutlass::PipelineState<DispatchPolicy::Stages>;
 
   using PipelineParams = typename MainloopPipeline::Params;
+
+  // One threads per CTA are producers (1 for operand tile)
+  static constexpr int NumProducerThreadEvents = 1;
+
   using ScaleTileShape = decltype(make_shape(shape<0>(TileShape{}),
                                              shape<1>(SmemLayoutAtomScale{})));