[Performance, Triton Kernel Args] _decode_grouped_softmax_reducev_fwd…

[HaiShaw]

… speedup on ROCm

Motivation

Speedup _decode_grouped_softmax_reducev_fwd.
Test shows ~1.0% improvement to median decode throughput on MI300x with Grok-1 and FP8 (b32/i1024/o256)

Modifications

Setting optimal kernel arguments to _fwd_grouped_kernel_stage2 on ROCm.

Checklist

• [+] Format your code according to the Contributor Guide.
• [+] Add unit tests as outlined in the Contributor Guide.
• [+] Update documentation as needed, including docstrings or example tutorials.