Fix CK 2stages MoE (always use BK1 = 16)

[ex-rzr]

Motivation

Fix for AICK-1084: FP8 2stages CK MoE produces incorrect results for intermediate dim < 256

Technical Details

It is incorrect to pass AK1Value = BK1Value = 8 to work around static asserts triggered in thread transfers for some configs.
Preshuffling (shuffle_weight) uses KPack = 16 for FP8, so BK1 must be the same otherwise the kernel loads wrong values.
Instead other parameters should be decreased in such cases:

• A/BBlockTransferThreadClusterLengths... so not all threads participate in loading (this commit)
• or A/BBlockTransferSrcScalarPerVector... so each thread load less data.

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:sglang	SGLang integration tests
ci:atom	ATOM benchmark (DeepSeek-R1 + GPT-OSS)
ci:vllm	vLLM benchmark
ci:all	All of the above

Add labels via the sidebar or gh pr edit 2898 --add-label <label>