use dequantize per channel group for embedding (#2374)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/2374

- add type hints to quantized_ops
- use dequantize per channel group in embedding byte decomposition

bypass-github-export-checks

Reviewed By: mikekgfb

Differential Revision: D54813256

fbshipit-source-id: 79b8f39d820378faa90d908e3ea56d4201a61598