G2 S2b: remove X5 Retail Hero loader, retire at-scale benchmark, de-brand repo-wide

[brownjuly2003-code]

Summary

Executes the S2a decision doc's (x5-benchmark-decision.md, deleted in the
final commit) disposition list in full, per the project owner's override:
the "X5 Retail Hero" dataset (a real external grocery retailer's public
Kaggle dataset) must not appear anywhere in this project going forward,
except as a historical mention in CHANGELOG.md.

• Delete warehouse/agentflow/dv2/loaders/x5_retail_hero/ entirely (no
  successor generator) and its dedicated test. Vault-generic row models
  needed by PostgresVaultWriter's only test coverage
  (test_dv2_postgres_ingestion.py) are relocated to a neutral
  loaders/vault_rows.py.
• Retire the at-scale load-test benchmark ("45.8M line items / 8.06M
  orders", 2026-06-07) as historical. The harness
  (infrastructure/dv2/load-test/) stays fully runnable and gating against
  the synthetic demo seed; docs/dv2-multi-branch/load-test-baseline.md is
  rewritten to the seed-scale baseline + a "Retired at-scale capture"
  section (full report preserved in this file's git history).
• Surgical test fixes: test_dv2_supplier_reference.py's X5-loader
  hash-equality pin → self-contained known-vector MD5 digests;
  test_dv2_business_vault_ddl.py / test_dv2_postgres_ddl.py drop the
  negative "x5__" not in DDL guard (the literal string would itself
  violate the no-X5-anywhere rule) and keep the positive "mp__" in DDL
  check.
• De-brand spec.yaml ("1c (X5 Retail Hero)" → "1c (branch order feed)") and regenerate its 10 dependent satellite DDL files.
• De-brand every remaining "X5" / "Retail Hero" / at-scale row-count
  mention repo-wide — dbt config, docs/domain.md §5.4,
  schema_dv2.md, demo_evidence.md, generator-spec.md, ops/perf docs,
  clickhouse-sts.yaml, dbt-run-job.yaml, business_vault files, DV2
  loader/reference Python docstrings. Several of these files were not in
  the original G2 audit's or the S2a decision doc's file list (dependency
  discoveries — see judgment calls below).
• CHANGELOG [Unreleased] entry added — the one sanctioned place
  "X5 Retail Hero" may still appear.
• x5-benchmark-decision.md deleted in the final commit (its own
  disposition, once executed).

Verification

• ruff check src/ tests/ scripts/ — clean
• ruff format --check src/ tests/ scripts/ — clean (361 files)
• mypy src/ --ignore-missing-imports — clean (124 files)
• pytest targeted (test_dv2_postgres_ingestion.py,
  test_dv2_supplier_reference.py, test_dv2_business_vault_ddl.py,
  test_dv2_postgres_ddl.py) — 105 passed
• pytest tests/unit/ tests/property/ (broad) — 1659 passed, 2 failed;
  both failures (test_sdk_wheel_contains_py_typed_marker,
  test_distribution_version_matches_sdk_version) are pre-existing local
  environment gaps (missing hatchling build backend, stale installed SDK
  version) in files this PR never touches (sdk/, tests/unit/test_sdk_packaging.py,
  tests/unit/test_version.py) — unrelated to this change.
• Grep-clean (git-tracked files only): x5 case-insensitive outside
  CHANGELOG.md returns only expected false positives — docker-compose.e2e.yml
  (-Xmx512m), docs/operations/aws-oidc-setup.md (X509...), and 4 binary
  screenshots/videos with coincidental byte matches. 45.8 / 8.06M /
  402K / "Retail Hero" outside CHANGELOG: clean except one legitimate
  45.8 QPS value in load-test-baseline.md's seed-scale results table
  (not a row count; copied verbatim from the decision doc's ready-made
  content).
• Rebased cleanly onto origin/main (dc23318, S1's currency/locale sweep,
  PR fix(g2-s1): sweep residual USD/generic-catalog text to kitchen/RUB legend #155) — zero file overlap, all checks re-run post-rebase.

Judgment calls worth a sanity check

• docs/dv2-multi-branch/schema_dv2.md's "Связь с реальными source-data"
  table previously pointed at three external Kaggle datasets (X5 Retail
  Hero + Lenta BigTarget); rewrote it to point at the synthetic seed / mp__
  feed instead of just deleting the X5 rows, since the table's whole premise
  (external prototype-data provenance) no longer describes the project.
• Several files outside the S2a decision doc's list needed de-branding
  (clickhouse-sts.yaml's benchmark-tuning comments, dv2/README.md's
  hash-key gotcha pointer, generator-spec.md's stale gs1.py docstring
  note, the two dated docs/perf/vault-pii-governance-pg-verify-2026-07-0{2,3}.md
  transcripts, postgres_oltp/README.md and reference/README.md's loader
  mentions) — treated the same as the decision doc's explicit list (keep
  the engineering rationale/meaning, lose the brand and exact at-scale
  numbers).
• load-test/README.md's scenario table had two other stale issues beyond
  the decision doc's ask (filenames missing _adhoc suffix, and
  05_line_items_reach's description still said uniqExact when the actual
  query already uses uniq()) — fixed both while touching that table.

🤖 Generated with Claude Code

[github-actions]

DORA Metrics

• Window: last 30 days
• Branch: main
• Deployment frequency: 83 total / 19.37 per week
• Lead time for changes: avg 0.56h / median 0.0h
• Change failure rate: 56.63% (47/83)
• MTTR: n/a across 0 incident(s)

MTTR note: No failed mainline CI runs in the selected window.