Missing ETP Support in mca_adapter Under Expert Parallelism

[vincent-hyx]

Could we consider adding support for Megatron-MoE's ETP (Expert-Tensor Parallel) sharding?
Right now, when initializing expert-parallel in MoE, it inherits the non-MoE tensor-parallel group by default.

[HuangJoJo]

This feature is on our v0.2 version roadmap！ Also Welcome to collaborate and submit your pull requests