Allow TruffleRuby CI to fail without failing the whole job

[rwstauner]

refs https://github.com/Shopify/yjit-bench/pull/332#issuecomment-2377505433

While this appears to work, it may make it hard to notice when the TruffleRuby build is failing (if the individual test looks green):

[eregon]

Switching to a TruffleRuby release seems better, because indeed this is completely hiding the fact the job fail and e.g. a breaking change in the harness would be unnoticed. (https://github.com/eregon/truffleruby-gem-tracker would still notice the failure, but not PRs on this repo)

Alternatively, IMO it's also fine to keep things as-is, the job only started to fail when updating protoboeuf in #325. Skipping the benchmark and pinging me sounds a reasonable workflow to me, that doesn't seem to be much overhead for yjit-bench maintainers, is it?

[eregon]

To get some data I looked at failed runs at https://github.com/Shopify/yjit-bench/actions/workflows/test.yml?query=branch%3Amain In the first 5 pages, all red jobs seem to be when ruby-head was failing, not truffleruby-head, except https://github.com/Shopify/yjit-bench/actions/runs/11002156114/job/30548468755 where they were both failing for the same reason. IOW ruby-head seems more unstable than truffleruby-head recently, so I think there is no need to change anything here.

[k0kubun]

that doesn't seem to be much overhead for yjit-bench maintainers, is it?

To me, it does. Nothing feels like more of an overhead than false positives for the repository.

Switching to a TruffleRuby release

I'm open to trying this first too. As long as the job doesn't fail unless we bump the TruffleRuby version or we push CRuby-specific code, I'm good.

[eregon]

Let's do https://github.com/Shopify/yjit-bench/pull/334 then