Update Cutlass to v2.11 (#94188)
Now that we are on CUDA 11+ exclusively, we can update Nvidia's Cutlass to the next version. We also had to remove the cuda build flag : "-D__CUDA_NO_HALF_CONVERSIONS__" since Cutlass no longer builds without it.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94188
Approved by: https://github.com/ezyang, https://github.com/jansel
