Fix quantized avg_pool2d test to support non-zero padding (#28246)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28246
Updated the reference fp32 implementation to use the dequantized input tensor to correctly take padded values into account
Test Plan:
python test/test_quantized.py TestQNNPackOps.test_avg_pool2d
Imported from OSS
Differential Revision: D17989334
fbshipit-source-id: 848ce78713280f529f71ff48e930db8de18abc62
