Increasing precision for avg pool (#23906)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/23906
The 1-off error is expected for the average pool due to double rounding.
Increasing unittest precision tolerance to 1.0 to avoid failing.
Test Plan: Imported from OSS
Differential Revision: D16678044
Pulled By: zafartahirov
fbshipit-source-id: 4e73934e4379b1d108af649ec77053998e44c560