### What this PR does / why we need it? 1. Move ops/comm_utils to ops/moe/comm_utils 2. Move distributed/tensor_parallel/gather_from_sequence_parallel_region to ops/moe/comm_utils 3. Delete distributed/tensor_parallel ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? e2e & ut - vLLM version: main - vLLM main: https://github.com/vllm-project/vllm/commit/a1213fae5f8fe624c4b954768d23a86e384abdb1 --------- Signed-off-by: wuweiqiang24 <1005334931@qq.com> Signed-off-by: wuweiqiang24 <wuweiqiang11@huawei.com>