[AMD] Skip the AITER #6042/#5967 patches on gfx1250

[michaelzhang-ai]

What

Apply the ROCm/aiter#6042 cherry-pick and the ROCm/aiter#5967 FMoE CSV diff only when GPU_ARCH_LIST != gfx1250.

Why

#42023 added both patches to the shared AITER clone step, but gfx1250-rocm1000 still pins AITER a6d2b564, which is 412 commits behind the e7d2453f pin the patches were written for. Since then the nightly publish-gfx1250 job fails (run 36861890479):

CONFLICT (content): Merge conflict in aiter/ops/triton/attention/pa_decode_sparse.py
error: could not apply 042a28918... Avoid the peeled sparse MLA spill on 64-bit gathers below triton 3.8

Fixing only the cherry-pick would not be enough: the #5967 diff modifies dsv41_fp4_{tuned,untuned}_fmoe.csv, which do not exist at a6d2b564, so git apply fails next.

Notes for the reviewer

The gate uses GPU_ARCH_LIST, like the other gfx1250-specific paths in this file. gfx942/gfx950 are unchanged. Both patches target DSV4.1 paths that gfx1250 did not have before #42023 either.

Test plan

Ran the clone/patch/revert part of this RUN step (without submodule update and pip install) against a local AITER clone:

• gfx950 @ e7d2453f: cherry-pick and CSV apply succeed, same as before.
• gfx1250 @ a6d2b564: patches skipped, sed and the four gfx1250 reverts apply cleanly.
• gfx1250 @ a6d2b564 with the current block: reproduces the cherry-pick conflict and the missing-CSV git apply error.

Full publish-gfx1250 build of this branch via workflow_dispatch succeeded: run 36999577158. The AITER step skips both patches and the four gfx1250 reverts apply.