Int8FC dequantize fix (#40608)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40608
Changes to fix uint8_t to fp16 dequantization error.
Enabled test_int8_quantize
(Note: this ignores all push blocking failures!)
Test Plan: Verified with test_int8_ops_nnpi.py
Reviewed By: hyuen
Differential Revision: D22252860
fbshipit-source-id: bb44673327f0c8f44974cef2ab773aa0d89f4dc7