[llvm] [LTO] Fix a crash with thin LTO caching and asm output (PR #138203)

[Alexey Karyakin via llvm-commits]

quic-akaryaki wrote:

> > > Is this fall out of #136121? If so, I wonder if that PR should be reverted instead - it turns out that went in without a full review / approval, and if it is causing subtle bugs I would prefer to revert that one.
> > 
> > 
> > I don't have a strong opinion, someone has to make a decision. If #136121 is reverted, I would like to keep the test.
> 
> Keeping the test is good. What is the failure mode of the test without your fix? I've asked for a revert of PR136121 as it was supposed to make error handling easier, but here it sounds like there was a much more subtle issue produced by that change.

Without the "fix", the test crashes.

https://github.com/llvm/llvm-project/pull/138203