[c10d] fix nccl gather outputs on non-root ranks (#75535)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/75535
we need to fill a empty tensor to outputs list bc collective macro requires it
ghstack-source-id: 153443855
Test Plan: wait for ci
Reviewed By: fegin, fduwjj
Differential Revision: D35508587
fbshipit-source-id: e42a0baeec8a3ebccd5c9e7da3cb216ce64f28fd
(cherry picked from commit 2bef237f3825e37bc27480baa36849203d18faa6)
