[ci] do not continue through error on trunk (#65503)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65503
There are two reasons for this change:
- I don't think trunk jobs should have different behavior than their PR equivalents.
- Continuing through error makes it challenging to figure out what is
actually failing, especially given the poor UX of GitHub Actions when it
comes to reading logs
Example: https://github.com/pytorch/pytorch/runs/3680114581. Here, there
is a failure but the rendered test results tell me everything is
successful. I have no idea how to quickly tell what failed; the log is so long
and terms like "error", "failure", etc. are common enough that searching
it is very difficult.
Differential Revision:
D31130478
D31130478
Test Plan: Imported from OSS
Reviewed By: ezyang
Pulled By: suo
fbshipit-source-id: 15a80475ca4c49644c0f7b779f5c6c2ffeb946a6