OPS-125 M-DEADSEAT: distinguish NEVER-EXISTED from LAPSED dead seats

pmmaster finding 2026-09-12 (OPS-125 follow-up, credited to a live-roster check I ran reconciling mba-l names): db03-mba-l-cc has no -cc sibling anywhere - only db03-mba-l-ca exists and ever existed. It was escalated to Elazar as a dead/lapsed seat holding 42 rows incl. 5 p0s, but it never was a seat - the -ca-to-cc migration wrote a name with no counterpart to be a twin OF. Different remedy from a real lapsed seat: LAPSED (a -ca/-cc sibling exists, or the raw name was itself once registered) gets relaunched; NEVER-EXISTED (no agent of that name in the roster, ever, and no sibling of the same stem under the other suffix) gets its rows REPOINTED to whichever sibling actually exists - nobody will ever pick them up by accident otherwise. M-DEADSEAT currently calls both cases the same 'dead seat' label, which sends the reader at the wrong fix. Add a third state to its output: NEVER-EXISTED vs LAPSED, alongside the existing LIVE PROJECT DEAD SEAT / dormant backlog tag. Cheap - same roster/namespace data already computed, just check whether ANY -ca/-cc sibling of the dead name's stem was ever registered (roster's own historical registered_at exists per-agent; a name absent from the CURRENT roster with no sibling present either way needs one more roster read, not new infra). Deferred behind KPI-69 (maintainer-registry signal), file order not urgency order.