Add the MegaMoE A2A backend name to server argument choices and MoE backend enums, wire the startup handling that forces ep_size to tp_size, and add the env flags used by the later MegaMoE forward and weight layout work. Treat MegaMoE as a dispatcher bypass so it does not enter DeepEP dispatcher logic.\n\nConstraint: Applies on top of origin/rc/v0.1 for GLM 5.2 MegaMoE adaptation.\nFeature-flag: --moe-a2a-backend megamoe and SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE.\nConflict-hotspots: python/sglang/srt/server_args.py; python/sglang/srt/layers/moe/utils.py; python/sglang/srt/layers/moe/fused_moe_triton/layer.py\nScope-risk: medium\nTested: git diff --check\nTested: python3 -c ast.parse on modified Python files
beb2162028