Revert "[Performance] Move apply_w8a8_block_fp8_linear to an op class… (#25607)

Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>

benchmarks/kernels/deepgemm/benchmark_fp8_block_dense_gemm.py

@@ -9,7 +9,7 @@ import torch
 from vllm import _custom_ops as ops
 from vllm.model_executor.layers.quantization.utils.fp8_utils import (
     per_token_group_quant_fp8,
-    w8a8_triton_block_scaled_mm,
+    w8a8_block_fp8_matmul,
 )
 from vllm.triton_utils import triton
 from vllm.utils.deep_gemm import (
@@ -63,7 +63,7 @@ def benchmark_shape(m: int,
 
     # === vLLM Triton Implementation ===
     def vllm_triton_gemm():
-        return w8a8_triton_block_scaled_mm(A_vllm,
+        return w8a8_block_fp8_matmul(A_vllm,
                                      B_vllm,
                                      A_scale_vllm,
                                      B_scale_vllm,