[Bug] composable_kernel submodule is missing

[xmfbit]

In the last commit, you add a new 3rdparty composable_kernel, but not add it in the submodule, which will cause the compilation fail.

I find that composable_kernel is a ROCm repo. So I just remove it, and it works well on CUDA.

[LeiWang1999]

Thank you for reporting this issue, @xmfbit! this is indeed a bug, as the composable kernel has been included within the 3rdparty/tvm. Could you please submit a pull request to remove the composable kernel include in setup.py? That would be greatly appreciated!