test: Compiler: relax flaky codegen test

[nsajko]

Relaxing it should help prevent spurious CI failures until someone does a proper fix.

Updates #56748

[adienes]

why not just @test_broken that seems harder to forget about (vs this approach seems easy to accidentally never re-tighten the bound)

[giordano]

Test broken shouldn't be used for tests which fail intermittently, you'd get test failures the tested condition is true. You can skip it, but marking as broken isn't correct.

[adienes]

agreed, @test_skip looks more appropriate than @test_broken

[Keno]

it would still be good to understand where this failure is coming from and if there is a sequence of tests that if run on the same processor deterministically causes this failure. Usually what I do is JULIA_CPU_THREADS=1 runtests.jl ... and then the list of all the tests that ran on the same worker on CI.

[oscardssmith]

Is this unrelated to the fact that @allocated on nightly is broken?

[nsajko]

Is this unrelated to the fact that @allocated on nightly is broken?

At least part of the issue here seems to be system-specific, so I guess it's a codegen issue 🤷