[quant] Add non-fbgemm fallback implementation for embedding lookup ops (#50706)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/50706
Add a default CPU implementation for quantized embedding lookup operators.
This should enable the ops to execute on mobile as well where we don't have fbgemm.
Test Plan:
python test/test_quantization.py
and CI tests
Imported from OSS
Reviewed By: vkuzo
Differential Revision: D25956842
fbshipit-source-id: 07694888e5e1423b496af1a51494a49558e82152