[CI/Build] Conditionally register cutlass_fp4_group_mm to fix building on Hopper (#26138)

Signed-off-by: mgoin <mgoin64@gmail.com>

csrc/torch_bindings.cpp

@@ -397,7 +397,7 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
       " Tensor a_blockscale, Tensor b_blockscales, Tensor alphas,"
       " Tensor problem_sizes, Tensor expert_offsets, Tensor sf_offsets) -> ()",
       {stride_tag});
-  ops.impl("cutlass_fp4_group_mm", torch::kCUDA, &cutlass_fp4_group_mm);
+  // conditionally compiled so impl registration is in source file
 
   // CUTLASS w8a8 GEMM, supporting symmetric per-tensor or per-row/column
   // quantization, as well as bias