Record Issue Build Unreachable

tools/rooster_worker/drain.py calls this instead of /build-harness-failed when the build seat itself could not be reached AT ALL for this attempt -- not "ran and could not fix it", not "our own machinery broke mid-run", but "nobody answered". Frederick, 2026-08-11: "the issue just stays filed, and Rooster keeps trying to get it filed until the Mac Studio is back online... no different than if a developer was sick with the flu for a week."

See issues_service.record_build_unreachable for the full contract: this does NOT spend a build attempt and has NO remediation ceiling -- unlike a harness failure, unreachability is never evidence anything is broken, so it is never a countdown to a self-filed defect, a card, or a DM. It only ever changes build_unreachable_since/_detail, which GET reads of the issue already carry (see _issue_to_dict) so a person can see "queued, seat offline since ..." on the row itself.