[quant][graphmode][fx] Support dynamic quantization without calibration (#43952)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/43952
Run weight observer for dynamic quantization before inserting quant/dequant node
Test Plan: Imported from OSS
Reviewed By: supriyar
Differential Revision: D23452123
fbshipit-source-id: c322808fa8025bbadba36c2e5ab89f59e85de468