[omni, model, config, docs] feat: support Qwen3-MoE in SeedOmni V2

Modules (6 files, +351): `qwen3_moe/llm/`, `convert_model.py`. Configs (6 files, +163): `configs/seed_omni/Qwen/qwen3_30b_a3b/`.

Reviewer focus: EP `parallel_plan`, and `ep_sharded_stream_load` enabled on this module alone — the per-module accelerator override from #1060 is what makes that possible in a heterogeneous omni model.

Start by inspecting the existing module structure under qwen3_moe/llm/ and convert_model.py, then compare the configs under configs/seed_omni/Qwen/qwen3_30b_a3b/. Review the per-module accelerator override from #1060, with particular attention to EP parallel_plan and ep_sharded_stream_load. Done means the Qwen3-MoE module and its six configuration files support SeedOmni V2 as described.

Written by the indexing model from the issue text.