Fix grouped GEMM with CUTLASS DSL 4.7 (#3369)

The Hopper grouped GEMM wrapper selected the current NVVM TMA bindings only when the CUTLASS version prefix was exactly 4.6. A 4.7 wheel therefore fell back to the removed CpAsyncBulkTensorLoadMode/loadMode interface.

Detect the TMALoadMode capability directly, preserving the legacy fallback for older bindings. For example, 4.7 now uses the TMALoadMode/mode path and descriptor address-space cast.

Verification: Python syntax compilation and diff checks.

examples/python/CuTeDSL/cute/hopper/kernel/grouped_gemm/grouped_gemm.py

@@ -121,9 +121,9 @@ def CpAsyncBulkTensorGlobalToSharedClusterOpWrapper(
     loc = None,
     ip = None,
 ) -> None:
-    if cutlass.__version__[0:3] == "4.6":
+    if hasattr(_nvvm_d, "TMALoadMode"):
         from cutlass._mlir.dialects import llvm as _llvm_d
-        # Handle new nvvm API in 4.6+
+        # Handle the current NVVM API when available.
         tmaDescriptor = _llvm_d.addrspacecast(
             _llvm_d.PointerType.get(_CuteAddressSpace.generic), tmaDescriptor, loc=loc, ip=ip
         )