nx_store_janitor_beat.nx

nx_store_janitor_beat.nx -- standing reclamation of SUPERSEDED seg-store segment files. WHY A BEAT AND NOT A ONE-SHOT: on 2026-07-30 a single manual sweep moved 57,348 MB across 336 planes (14,585 live segment .docs down to 2,973). Within the SAME session debt- went from 1 segment back to 18 and down again as siblings wrote and compacted. The seg-store is append-only and every commit writes a NEW segment, so superseded files re-accumulate CONTINUOUSLY. A cleanup that has to be remembered is a workaround; the fix is a standing sweep row that keeps the property true without anyone noticing. SIBLING TO nx_store_fold_beat AND DELIBERATELY ORDERED AFTER IT: fold_beat MERGES a bloated plane (segment COUNT / scan cost), which by construction supersedes the segments it merged. This beat then reclaims those files (disk / directory hygiene). Fold makes garbage; janitor takes it out. Neither replaces the other. Forks the proven ./nx_store_janitor.elf per plane exactly as fold_beat forks nx_store_compact -- the worker owns the safety argument (manifest IS the definition of live, REFUSES on an unreadable manifest, moves aside into retired/ and never unlinks, and re-reads the manifest afterwards to restore anything a concurrent writer made live mid-walk). This beat adds discovery and aggregation ONLY; it must not re-implement policy. Sweep-row contract: exit 0 = every plane reclaimed or already clean; exit 1 = a plane REFUSED (unknown live set) or a live segment was moved and restored -- surfaced RED, never silent. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0