TODO.md cleanup: remove SHIPPED rows, refresh stale refs, add tiered backlog

Wave 1 of multi-wave tech-debt paydown. Docs-only — no estimator code touched.

Changes:
- Delete 3 SHIPPED rows marked `Done`: HAD trends_lin (PR igerber#389),
  HAD yatchew_hr_test(null="mean_independence") (post-PR igerber#392),
  HAD Phase 5 follow-up tutorial T22 (PR igerber#440).
- Fix stray blank line inside Methodology/Correctness table (was rendering
  as two fragments).
- Correct dCDH survey cell-period allocator row's PR-column "PR 2" → "igerber#408"
  (actual dcdh-by-path-survey-design merge).
- Refresh Known Limitations line refs to current state of estimators.py
  (`:778-784` → `:1647`; `:567-588` → `:890-911`).
- Regenerate Large Module Files table from `wc -l diff_diff/*.py` ≥1000 lines.
  Add 11 newly-eligible modules (chaisemartin_dhaultfoeuille.py 8636,
  had_pretests.py 4951, had.py 4593, etc.). Update header sentence to make
  the 3000/2000/1000 thresholds explicit instead of silently relaxing the
  documented "< 1000 lines" target.
- Rewrite Standard Error Consistency section — `vcov_type` has subsumed the
  proposed `se_type` knob; point readers at the existing methodology row
  for the remaining 8-standalone-estimator threading work.
- Modernize Test Coverage section to reference `pytest.importorskip` rather
  than a stale "21 visualization tests" count.
- Add new `### Prioritized Tech-Debt Backlog` subsection: Tier A/B/C/D
  ordering by effort × risk, anchoring back to existing source-of-truth
  table rows so the tables remain canonical. Seven Tier A quick wins,
  twelve Tier B mid-size methodology items, nine Tier C derivation items,
  eight Tier D deferred/research items.

The new tier structure exists so subsequent waves can be picked off
without re-litigating priority each turn. Wave 2 candidate is the
clubsandwich_cr2_golden.json regeneration from R (row 87 / Tier A).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: igerber