fix(ci): closure_gate.py seed_phi column missing on ssot.bpb_samples

[gHashTag]

Closure Gate (#446 — 312/312) failing on seed_phi column

Symptom: Closure Gate (#446 — 312/312) workflow has failed on every run since SHA dc02f63 (4 consecutive scheduled runs, latest 25639181566 / 2026-05-10T20:37Z).

Root cause: .github/scripts/closure_gate.py line 127 references seed_phi column but the live ssot.bpb_samples table on Railway / Neon does not have that column.

psycopg2.errors.UndefinedColumn: column "seed_phi" does not exist
LINE 1: ... MIN(bpb), COUNT(*), COUNT(DISTINCT seed_phi), MAX(step) FROM ssot.bpb_samples
                                              ^

Producer side (scripts/run_priority_matrix.ts + scripts/postrun_commit_back.ts) writes seed_phi to assertions/matrix_samples.jsonl (R7-honest local ledger), but the Railway/Neon ingest path stores it under a different name (likely seed or seed_value).

Three remediation paths:

1. DDL migration (preferred per R6) — add seed_phi INT column to ssot.bpb_samples, backfill from seed. Owned by trios-railway#62.
2. Producer change — rename seed_phi → seed in closure_gate.py query if the canonical column is seed.
3. R5 honest disable — mark Closure Gate as continue-on-error: true until the SoT is reconciled.

Affected runs (all event=schedule, head_sha=dc02f63):

• 25634067779 / 25635358071 / 25636663465 / 25638024080 / 25639181566

R-rules: this is NOT attributable to Wave-29/31/32 SHAs (those merged on gHashTag/trios-trainer-igla, not gHashTag/trios). Coq-runtime-invariants v1.1 attribution rule excludes pre-existing main breakage; Apiary cron NEW-CI-failure alert fired because dc02f631 is a NEW SHA on main but the breakage is schema-drift, not Wave-32 regression.

Anchor: phi^2 + phi^-2 = 3 · DOI 10.5281/zenodo.19227877

cc trios-railway#62 (bpb_samples DDL drift), trios#446 (closure parent)