[Bugfix] Register fp8 cutlass_group_gemm as supported for only SM90+SM100 (#33285)

Signed-off-by: mgoin <mgoin64@gmail.com>
(cherry picked from commit 1bd47d6e5a)

vllm/_custom_ops.py

@@ -900,6 +900,8 @@ def cutlass_sparse_scaled_mm_supported(cuda_device_capability: int) -> bool:
 
 
 def cutlass_group_gemm_supported(cuda_device_capability: int) -> bool:
+    if cuda_device_capability < 90 or cuda_device_capability >= 110:
+        return False
     try:
         return torch.ops._C.cutlass_group_gemm_supported(cuda_device_capability)
     except AttributeError: