Add support for native FP4 Tensor Core MMA in build script

Author: czxdev

.github/workflows/build-wheels-cuda.yaml

@@ -227,6 +227,9 @@ jobs:
             # CUDA 13 dropped offline compilation support for pre-Turing targets.
             $cudaArchs = "75-real;80-real;86-real;89-real;90-real;90-virtual"
           }
+          if ([version]$nvccVersion -ge [version]"12.8") {
+            $cudaArchs = "$cudaArchs;120a-real"
+          }
           # Build real cubins for the supported GPUs and keep
           # one forward-compatible PTX target instead of embedding PTX for every
           # SM. This keeps the wheel under GitHub's 2 GiB release-asset limit.