corpus_unresolved: each_backend_artefact_write_is_guarded_at_its_own_call_site is stale since #4733 (c.rs -> c-<pid>.rs) #5611

Visible since #5455 (c7acbe5) made cargo test compile again.

cargo test --release --offline --test corpus_unresolved gives 6 passed and 1 failed. each_backend_artefact_write_is_guarded_at_its_own_call_site panics at bootstrap/tests/corpus_unresolved.rs:309: "an unwritable c.rs must refuse the run on its own", because the run exits 0.

Cause: #4733 (7176266) renamed the corpus artefacts to per-process names (bootstrap/src/service.rs: tmp.join(format!("c-{}.rs", std::process::id())), and likewise for .c/.zig/.v). The test still pre-creates a DIRECTORY named c.rs / c.c / c.zig / c.v under TMPDIR/t27-corpus, so the EISDIR trap never fires. The test went stale while the target could not compile.

Fix options: have the test learn the child's artefact names (for example an env override for the artefact stem, or a --artefact-dir), or make the trap a non-writable t27-corpus subdirectory per backend. The guard itself (service.rs ~1227) still exists; only the test lost its aim.

Found by reviewer bee W.

Boundary

• bootstrap/src/service.rs
• bootstrap/tests/corpus_unresolved.rs