Compliance: scans that never finish leave the matrix

Found while photographing the Compliance screens on prod (docs-writer Task 6602). In one org, all four projects have read "Scanning" with their Auditor run links since 2026-10-13 (four days), the matrix reads "0 of 4 scanned · running", Rescan all stays disabled as "Scanning…", and Order is locked ("Sorting is available once the scan finishes"). Suggestions: 1. A scan whose auditor run has ended (or exceeded a time bound) without a result should turn that row into **Scan error** with **Open to rescan**, never stay "Scanning" indefinitely. 2. While a scan runs, offer a way to stop it (or rescan one stuck row) instead of leaving **Rescan all** disabled with no recourse. 3. A row that already has a level from a previous scan should say when that result is from while the new scan runs, so a stale level is not read as current. 4. Let **Order** sort the rows that already have a result even while others are still scanning. Out of scope: rule content, packs, the Architectural Guard, the docs pages. Done means: a scan whose auditor run ended without submitting turns its row into Scan error within a bounded time (test proving it); Rescan all is usable when every running row is stuck; tests cover each change; ./dc check full green.

Stuck compliance scans now turn into Scan error after 6h. Added Stop scan, which frees each run's slot, a "Result from" date on older results, and Order while a scan runs. Full gate green; PR #5599 merged, its deploy superseded by a newer one. Drafts filed: #6666 docs, #6669, #6681.