[GHF][BE] Memoize `read_flaky_rule (#101239)
Added caching to `read_flaky_rules`, as it's called several times during the merge process and every call incurs network access. Also, one should not expect flaky rules to change while PR is landed.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101239
Approved by: https://github.com/huydhn, https://github.com/kit1980
