bug: operator_queue.type is unbounded at ingest and at the DB sink #3385

Deferred-debt entry (verbatim from the trinity-dev DEBT_INBOX.md)

debt:2026-09-23-operator-queue-type-unbounded-at-ingest

What: operator_queue.type is agent-authored free text with no bound at ingest (_clamp_ingested_item bounds title/question/context/options/priority/created_at, not type) and no belt at the DB sink (create_item stores item.get("type", "question")); the only cap is the 2 MiB whole-file limit. The row is prunable/clearable and the frontend renders it as text through a label map, so the impact today is bloat, not control flow.
Why deferred: pre-existing (predates #2915); the #2915 audit found it only because the NEW ingested audit row copied the field into the immutable audit table — that copy is folded now. Bounding the column is a #1632-style clamp with a truncate-with-marker decision and a _BUDGETED_ALERT_TYPES-adjacent question (platform emitters mint non-protocol types on purpose), so it deserves its own small PR.
Context: services/operator_queue_service.py::_clamp_ingested_item; db/operator_queue.py::create_item (_DB_BELT_* cover id/title/question/context); utils/operatorQueue.js::queueTypeLabel renders the raw value as a fallback; the WS operator_queue_new payload carries it.
Interest: an agent can put a multi-megabyte string on a queue row and its broadcast; every reader that echoes type (label fallback, MCP list) carries it.
Effort (AI): S (clamp + belt + one test).
Source: #2915 — cso --diff report F1 residual (verified by the independent verifier, 2026-09-23).

Filed by trinity-pm /groom 2026-10-08 (Debt Health & Inbox Triage).