pytorch

80150788 - [21/N] Add alltoall_base custom op with CPU/CUDA implementations (#89813)

[21/N] Add alltoall_base custom op with CPU/CUDA implementations (#89813) Differential Revision: [D41812670](https://our.internmc.facebook.com/intern/diff/D41812670) Pull Request resolved: https://github.com/pytorch/pytorch/pull/89813 Approved by: https://github.com/kwen2501