🧪 test(coverage): remove three lines no run can execute

[gaborbernat]

main has failed its own coverage gate since #2178, so nothing can merge. #2192, #2187 and #2162 all report the same seven lines, and no diff among them touches one. This removes those lines.

git log -S on each of the three sites returns one commit, d75b16a12, which is #2178. None of them existed before it, so nothing earlier can have exposed or masked them, and no merge artefact is needed to explain the timing. The earlier failure at 0df149bfc was a different fault in the same job, a panic that reported no uncovered lines at all, and it is filed as #2200.

Two of the lines are validate_retention on the retention doubles #2178 added. The trait requires the method, so neither body can go away, and a default returning Ok(()) on a validation hook would beat the dead line only by hiding a driver that forgot to validate. What the two doubles can do is what the file's three older doubles do: validate the policy they were handed as the first act of plan_retention. #2178 added two that skip that step, leaving a method every driver must implement with no caller anywhere. Restoring the step gives it the same caller the others have, and the gated double keeps asserting exactly what it did.

The panicking double needed a different answer, which CI caught rather than my reasoning. Validating before the panic moved the dead line instead of removing it: a panic! never completes, so its region never records, and the entry region that had been covering that line moved up to the validation. Seven uncovered lines became one. That double now comes apart inside validate_retention, with plan_retention as the call that asks for it, so both required methods sit on the path the test drives and each is a single statement its entry region covers. The worker still dies inside the plan, which is what the test asserts.

The third is a straight deletion. published.is_ok() sat in an assert_eq! failure message, which evaluates only when the assertion trips, and the line above it already asked the same Result whether the publish failed. Binding the answer once before the comparison removes the unreachable evaluation and the second pass over the Result together.

Measured rather than reasoned. cargo llvm-cov nextest -p peryx-driver -p peryx-ecosystem-oci now reports neither tests/unit/retention/tests.rs nor tests/unit/quota/tests.rs in its uncovered list, and 1908 tests pass across the two crates.

Closes #2199

[codspeed-hq]

Merging this PR will not alter performance

✅ 29 untouched benchmarks
⏩ 133 skipped benchmarks¹

---

_{Comparing fix/uncoverable-test-lines-2198 (d46eba4) with main (2d7849a)}

Footnotes

1. 133 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩