Route fused_sample_helion_kernel

src/fused_mm_sampling/helion_impl.py:None–None

helion.kernel(autotune_effort=os.environ.get("HELION_AUTOTUNE_EFFORT", "quick"))

"""
Helion implementation for benchmarking purposes.
This module is separate from core to avoid making helion a required dependency.
It is only used for performance comparisons.