🧪 test(driver): close the jobs and rate-limit survivors

[gaborbernat]

A fresh cargo mutants -p peryx-driver pass over jobs/mod.rs and rate_limit.rs, 226 mutants, found the crate's own suite blind to three of them and unable to decide five more. This closes all eight for #1893, and both files now measure zero survivors. 🎯

Seven jobs built the same JobReport literal by hand, and the copies in the cancelled arms of the history cleanup, the ledger reap and the aborted search rebuild could drop either counter unnoticed. A cancelled sweep only reports work when cancellation lands between two store batches, which no caller can place, so those literals carried a decision nothing could observe. Two constructors, JobReport::changed for a run whose changed count equals its processed count and JobReport::examined for a sweep that changed nothing, replace the copies; the decision now lives in one place the succeeded arms already prove, and an aborted rebuild reads as the examined case it is.

The two drain loops gained the states that decide their exit. A history backlog wider than the store's 128-run batch shows the cleanup keeps pruning past the first batch, and a reap with one row per ledger per pass shows the loop runs until every ledger is empty rather than until the first one is; under the mutation the expiry ledger empties first and the loop leaves the outcomes for a later tick. ♻️

RateLimiter::trusts_any_proxy had no caller in the crate; peryx-http asks it before reading forwarded headers. A limiter answering a constant either way passed the suite, so one case now pins both answers. The two ActorKey::kind survivors this crate was tracking as blocked are already dead: #2213 ported the binary-lifetime capture subscriber and pinned the client kind a refusal names, and the run here catches both.

Refs #1893

[codspeed-hq]

Merging this PR will not alter performance

✅ 29 untouched benchmarks
⏩ 133 skipped benchmarks¹

---

_{Comparing test/driver-mutants-1893-b (d77311d) with main (a7dab14)}

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. ↩