[quant][fix] A typo in quantized::conv2d_relu (#37964)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37964
I thought it was because of flakiness that we didn't pass the conv2d_relu test, but turns out to
be a typo in the implementation
Also re-enabled the `use_fused` option in `test_conv2d_api`
Test Plan:
.
Imported from OSS
Differential Revision: D21434776
fbshipit-source-id: 7c24c13cde0a96807e8dfbd1deabf47e8280fdb7