Fix f8f8bf16_lite quantize op input in quantize_and_compute (pytorch#745)

Summary:
X-link: pytorch#3667

Pull Request resolved: facebookresearch/FBGEMM#745

A minor fix for trt-llm cudaCoreGemm `cuda_lite` op in quantize_bench script.

when testing with `--bench_quantize` detected a failure with input

```
...
tree/deeplearning/fbgemm/fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py", line 797, in quantize_and_compute
    return self.compute(xq, wq, x_scale * w_scale)
TypeError: FP8LiteGemm.compute() missing 1 required positional argument: 'w_scale'
```

Reviewed By: jwfromm

Differential Revision: D69272912

fbshipit-source-id: c184954b4d2d1543277a9e56ac899534597a56e6

Author: YUNQIUGUO

fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py

@@ -719,7 +719,7 @@ def compute(self, xq, wq, x_scale, w_scale):
 
     def quantize_and_compute(self, x, w):
         xq, wq, x_scale, w_scale = self.quantize(x, w)
-        return self.compute(xq, wq, x_scale * w_scale)
+        return self.compute(xq, wq, x_scale, w_scale)
 
     @property
     def name(self) -> str: