add conversion functions to embedding tables (#31083)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31083
add (fp32/fp16)<->(int8 rowwise quantized fp32/fp16 scale biases)
Test Plan:
added unit tests
enhanced shape inference tests
Reviewed By: jspark1105
Differential Revision: D18920547
fbshipit-source-id: 6b3d7cb93f9d1669ecf511817d73976177632891
