Fix CI after quantize op change in PyTorch core

Summary: pytorch/pytorch#125781 recently
changed the numerics of the quantize op subtly. This commit fixes
the numerics mismatch caused by this PR by making our quantize
ops consistent with the ones in core.

Test Plan:
python test/quantization/test_quant_primitives.py -k test_quantize_dequantize_group_sym
python test/quantization/test_quant_api.py TestQuantFlow.test_quantized_tensor_subclass_8da4w

Reviewers: jerryzh168, cpuhrsch

Subscribers: jerryzh168, cpuhrsch, supriyar

torchao/quantization/quant_primitives.py

@@ -201,7 +201,7 @@ def quantize_affine(
 
     if zero_point_domain == ZeroPointDomain.INT:
         quant = torch.clamp(
-            torch.round(input / scale) + zero_point, quant_min, quant_max
+            torch.round(input * (1.0 / scale)) + zero_point, quant_min, quant_max
         ).to(output_dtype)
     else:
         assert zero_point_domain == ZeroPointDomain.FLOAT