[MOE] Add fused dynamic MXFP8 quant + moe_sort HIP path

[yadaish]

Adds fused_dynamic_mxfp8_quant_moe_sort_hip that quantizes activations to fp8 with e8m0 group scales and writes the swizzled scale layout consumed by the FlyDSL a8w4 stage1/stage2 GEMM. Wires it into fused_moe_2stages to replace the Triton fused_quant_fp8_sort path, with AITER_MOE_A8W4_BYPASS_QUANT to fall back to the prior unquantized behavior.

For e8m0-scaled fp8 output, the divisor is switched to 1/floor_pow2(max) so row_scale stays a pure power of two and matches the 2^(byte-127) dequant scale encoded in the e8m0 byte.

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:triton-300x	Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang	SGLang integration tests: DeepSeek-R1-MXFP4 accuracy, Qwen 3.5 accuracy
ci:atom	ATOM benchmark: DeepSeek-R1-0528, GPT-OSS-120B
ci:atom_full	ATOM accuracy suite for PR and main models from ATOM models_accuracy.json
ci:vllm	vLLM benchmark: GPT-OSS-120B, DeepSeek-R1-0528, Kimi-K2.5
ci:all	All standard extended tests (excludes ci:atom_full)

Only add ci:atom_full for FlyDSL or Triton upgrades.
Add labels via the sidebar or gh pr edit 3312 --add-label <label>