[quant][embedding qat] Set FakeQuant zeropoint dtype matches observer (#68390)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68390
Observer zero_point's dtype can be float, in the specific case of `torch.per_channel_affine_float_qparams`.
This change sets FakeQuant's zero_point dtype accordingly.
Test Plan:
`pytest test/quantization/core/test_workflow_module.py -v -k "embedding"`
`pytest test/quantization/eager/test_quantize_eager_qat.py -v -k "embedding"`
Imported from OSS
Reviewed By: vkuzo
Differential Revision: D32446405
fbshipit-source-id: cca7aade68ff171887eeeae42801f77d934dad4c