Fix issue in s_addmm_out_sparse_dense_cpu only supporting CUDA device checking (#77018)
## Motivation
The at::native::s_addmm_out_sparse_dense_cpu only supports the CPU tensors. But it only checks whether the tensor is on CUDA device which is not enough.
## Solution
Change the tensor device type checkging from is_cuda to !is_cpu to protect other backends than the CUDA.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77018
Approved by: https://github.com/IvanYashchuk