fix(ci): Sprint 6 CI follow-up — staticcheck ST1021 + tenant-query baseline + skip inventory

Sprint 6 push (commits 43836ac + 663b14b) tripped three CI guards.
Fixing all three in this single follow-up — each is a small,
mechanical correction that doesn't change behavior:

1. staticcheck ST1021: AuditChainSnapshot doc comment was on the
   wrong type.

   internal/service/audit_chain_metric.go:91 had:
     // Snapshot returns the current counter state for the Prometheus
     // exposer. Reads use atomic loads — no mutex.
     type AuditChainSnapshot struct { ... }

   The comment described Snapshot() (the method on AuditChainCounter)
   but sat directly above the AuditChainSnapshot struct. staticcheck
   ST1021 requires exported-type comments to start with the type's
   name + optional leading article. Rewrote to lead with
   "AuditChainSnapshot is the point-in-time view ...".

2. multi-tenant-query-coverage: baseline drifted 31 → 32 because
   Sprint 6 COMP-002-RETENTION added UserRepository.ListDeactivatedBefore
   at internal/repository/postgres/user.go:191 — legitimately
   tenant-spanning by design.

   The retention policy is control-plane-wide (one
   CERTCTL_USER_RETENTION_WINDOW for the whole deployment, not
   per-tenant). The scheduler's userRetentionLoop walks every
   tenant's deactivated users on the same tick. A per-tenant
   tenant_id filter would require the scheduler to iterate every
   tenant — more code for equivalent semantics.

   Per the guard's own documentation (option b), legitimately
   tenant-spanning queries get an inline rationale comment + a
   baseline lift. Both delivered:
     - Inline comment block on the SELECT in user.go::ListDeactivatedBefore.
     - BASELINE_COUNT 31 → 32 in
       scripts/ci-guards/multi-tenant-query-coverage.sh, with the
       Sprint 6 rebase entry added to the rebase-history comment.

3. skip-inventory-drift: docs/testing/skip-inventory.md was stale.
   COMP-001-HASH added three new t.Skip sites in
   internal/repository/postgres/audit_chain_test.go (the three
   testing.Short() gates on the testcontainers integration tests).
   Re-ran ./scripts/skip-inventory.sh to regenerate the doc —
   totals went from 144 → 147 sites + 78 → 82 short-mode guards.

Verified locally:
  bash scripts/ci-guards/multi-tenant-query-coverage.sh      (clean)
  bash scripts/ci-guards/skip-inventory-drift.sh              (clean)
  go vet ./...                                                (clean)
  staticcheck ./internal/service/...                          (clean)

Closes the three Sprint 6 CI failures. The next CI run should
green out.

docs/testing/skip-inventory.md

@@ -8,8 +8,8 @@
 
 ## Summary
 
-- Total t.Skip sites: **144**
+- Total t.Skip sites: **147**
-- testing.Short() guards: **78** (these gate behind `go test -short`)
+- testing.Short() guards: **82** (these gate behind `go test -short`)
 
 Re-run inventory with: `./scripts/skip-inventory.sh`.
 
@@ -162,6 +162,9 @@ Re-run inventory with: `./scripts/skip-inventory.sh`.
 
 ### `internal/repository/postgres`
 
+- `internal/repository/postgres/audit_chain_test.go:137` — t.Skip("skipping integration test in short mode")
+- `internal/repository/postgres/audit_chain_test.go:36` — t.Skip("skipping integration test in short mode")
+- `internal/repository/postgres/audit_chain_test.go:58` — t.Skip("skipping integration test in short mode")
 - `internal/repository/postgres/audit_worm_test.go:29` — t.Skip("skipping integration test in short mode")
 - `internal/repository/postgres/auth_revoke_scope_test.go:118` — t.Skip("integration test in short mode")
 - `internal/repository/postgres/auth_revoke_scope_test.go:149` — t.Skip("integration test in short mode")

internal/repository/postgres/user.go

@@ -185,6 +185,13 @@ func (r *UserRepository) ListAll(ctx context.Context, tenantID string) ([]*userd
 // this list per tick and calls UserRetentionService.DeleteUserPII on
 // each. Cross-tenant on purpose: a single retention policy spans the
 // whole control plane.
+//
+// multi-tenant-query-coverage carve-out: the SELECT below intentionally
+// omits `tenant_id` because retention is a control-plane-wide policy
+// (one CERTCTL_USER_RETENTION_WINDOW for the whole deployment, not
+// per-tenant). Adding a `tenant_id = $N` filter would require the
+// scheduler loop to iterate every tenant, which is more code for
+// equivalent semantics. The guard's baseline counts this query.
 func (r *UserRepository) ListDeactivatedBefore(ctx context.Context, threshold time.Time) ([]*userdomain.User, error) {
 	rows, err := r.db.QueryContext(ctx,
 		`SELECT `+userColumns+`

internal/service/audit_chain_metric.go

@@ -88,8 +88,10 @@ func (c *AuditChainCounter) RecordBreak(brokenAtID string, brokenAtPos int) {
 	}
 }
 
-// Snapshot returns the current counter state for the Prometheus
+// AuditChainSnapshot is the point-in-time view of the counters the
-// exposer. Reads use atomic loads — no mutex.
+// Prometheus exposer reads. Snapshot() returns one of these; the
+// metrics handler renders each field into Prometheus exposition
+// format. Reads use atomic loads — no mutex required.
 type AuditChainSnapshot struct {
 	BreaksDetected uint64
 	WalksCompleted uint64

scripts/ci-guards/multi-tenant-query-coverage.sh

@@ -67,8 +67,16 @@ TARGET_DIR="${REPO_ROOT}/internal/repository/postgres"
 #
 # To rebase: re-run the guard, set BASELINE_COUNT to the new value,
 # include the rebase commit's SHA in the "last rebase" comment.
-BASELINE_COUNT=31
+BASELINE_COUNT=32
-# Last rebase: 2026-05-11 (Audit 2026-05-11 fix bundle dropped tenant_id-less
+# Last rebase: 2026-05-16 (Sprint 6 COMP-002-RETENTION added
+# UserRepository.ListDeactivatedBefore at internal/repository/postgres/user.go:191
+# — legitimately tenant-spanning by design. The scheduler's
+# userRetentionLoop walks every tenant's deactivated users on the
+# same tick; the retention policy is control-plane-wide, not
+# per-tenant. Documented inline in the SQL comment.
+#
+# Prior rebases:
+# 2026-05-11 (Audit 2026-05-11 fix bundle dropped tenant_id-less
 # queries by 1; v2.1.0 release-gate Phase 5 ratcheted baseline 32 -> 31).
 
 if [ ! -d "$TARGET_DIR" ]; then