Record Harness Failed Issue Build

tools/rooster_worker/drain.py calls this instead of /build-failed when the build's own machinery broke -- a timeout, a checkout that never started, CI that never reported, a merge or crash that has nothing to do with the fix -- rather than a build that ran, reasoned, and produced a wrong or failing change. See issues_service.record_harness_failure for the full contract: this does NOT spend a build attempt, and past a bounded remediation ladder it self-files a real defect against Rooster's own harness instead of ever landing "needs a person" on someone's Today for a fix nobody actually tried.

principal is passed through as BOTH actor_principal_id and filer_principal_id -- the calling agent's own authenticated identity is who touched the row and who a self-filed defect is authored by, exactly the same reuse record_issue_intake_attempt_failed above makes for the identical reason: never the reporter, who did not diagnose this.