Fix main whole-corpus red: nfr roster backfill for #6533's frontier_blocker_class_matches (selection-gap masking class, 3rd receipt)

[gunbai-bot]

Main is whole-corpus red: #6533 landed an unrostered non-fold-residue site

The red: non_fold_residue_clean_holds (dag witness) and non_fold_residue_no_unrostered_or_stale (lens witness) return false on current main — caught by a local whole-corpus falsifier cold sweep at 20:55 UTC (2 of 1,776 witnesses failing). Tonight's nightly falsifier will red on this.

Root cause, verified by execution: #6533 (Wave 2 frontier probe) added frontier_blocker_class_matches (src/v2/compiler/self_host/frontier_probe_types.dag) — five _ => wildcard arms on closed-coproduct params, the exact class the nfr census audits — without a roster row. The nfr witnesses are corpus-read, host-fed rows the affected-set selection did not run for that diff, so the landing PR stayed green and the red surfaced only on the next whole-corpus sweep. This is the third instance of the masking class receipted on #6530 (body-lowering claims, git-mock totality, now this) — each time a corpus-read witness's selection edge is missing, the red lands on the falsifier's doorstep instead of the offending PR's.

The fix (backfill precedent, 2026-07-12 rows): one dated roster row so the ratchet re-arms; burns down with the frontier probe's fold migration. #6533's other new wildcard fn (compiler_frontier_probe_entry_test.dag) is outside the nfr scan universe — deliberately no row (a dead row would trip the stale-roster count; verified: with it present, stale_roster_count=1).

Receipts: with this row — non_fold_residue_clean_holds PASS, non_fold_residue_universe_is_nonempty PASS, non_fold_residue_no_unrostered_or_stale PASS (claim_batch, live tree). Without it: unrostered=1.

Follow-up owed (not this PR): the missing-selection-edge class itself — corpus-read host-fed witnesses need never-predict-skip or an explicit selection edge; flagged on #6530, now with a third receipt.

🤖 Generated with Claude Code