Replace __host__ __device__ with C10_HOST_DEVICE in THCIntegerDivider.cuh (#39797)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39797
This change allows using [OffsetCalculator.cuh](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/cuda/detail/OffsetCalculator.cuh) on both CPU and CUDA.
It is required by (and manually tested with) https://github.com/pytorch/csprng/pull/28
Test Plan: Imported from OSS
Differential Revision: D22002571
Pulled By: pbelevich
fbshipit-source-id: 0b85e465777f6613f3b3ba91873da57dab949c54