docs(did): second backfill batch — 4 [@bibkey] annotations in did/ module

Second exercise of tools/suggest_bibkey_backfills.py. Cluster-picked
the did/ module's top-of-module docstring first lines — all 4 share
a uniform ``Title (authors YEAR, arXiv XXXX.YYYYY).`` pattern that
the suggester's score-3 filter identified as zero-risk end-of-reference
sites.

  cohort_anchored  -> [@cohort_anchored2025]   (arXiv 2509.01829, Liu 2025)
  design_robust    -> [@design_robust_es2026]  (arXiv 2601.18801, 2026)
  misclassified    -> [@did_misclassified2025] (arXiv 2507.20415, Augustin et al. 2025)
  did_bcf          -> [@wuthrich2025forests]   (arXiv 2505.09706, Wüthrich-Zhu 2025)

Each edit is a trailing ``[@key]`` append at the end of the first
prose period of the top-of-module docstring. No signatures, algorithm
paths, or function behaviour touched.

Verification
------------
  coverage:        64.6% -> 65.4%        (318 -> 322 cited keys)
  orphans:         174   -> 170          (4 keys fully consumed)
  candidate lines: 59    -> 55           (suggester re-run)
  candidate keys:  18    -> 14

  Gate 0  pytest 81 tests                                  pass
  Gate 1  audit_bib_duplicates --strict   0/0/0            pass
  Gate 2  audit_bib_coverage  --strict-dangling  0 dangling pass
  Gate 3  audit_citations     --strict    316/0/0          pass
  import test (4 touched did/ modules)                     pass

Cumulative over two batches (sensitivity_frontier + did/):
  coverage   64.0% -> 65.4%              (+1.4pp across 7 orphans)

Remaining: 14 backfillable orphan keys (55 candidate lines) —
reproducible with ``python3 tools/suggest_bibkey_backfills.py``.

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

Author: brycewang-stanford

src/statspai/did/cohort_anchored.py

@@ -1,5 +1,5 @@
 """
-Cohort-Anchored Robust Event Study (Liu 2025, arXiv 2509.01829).
+Cohort-Anchored Robust Event Study (Liu 2025, arXiv 2509.01829). [@cohort_anchored2025]
 
 Standard event-study TWFE with leads/lags is fragile under staggered
 adoption because untreated leads/lags from later cohorts contaminate

src/statspai/did/design_robust.py

@@ -1,5 +1,5 @@
 """
-Design-Robust Event-Study Estimation (arXiv 2601.18801, 2026).
+Design-Robust Event-Study Estimation (arXiv 2601.18801, 2026). [@design_robust_es2026]
 
 Provides design-consistent inference for TWFE event-study coefficients
 under staggered adoption. The key insight: under correct design (no

src/statspai/did/did_bcf.py

@@ -1,5 +1,5 @@
 """
-DiD-BCF — Forests for Differences (Wüthrich-Zhu 2025, arXiv 2505.09706).
+DiD-BCF — Forests for Differences (Wüthrich-Zhu 2025, arXiv 2505.09706). [@wuthrich2025forests]
 
 Non-parametric DiD via Bayesian Causal Forests (Hahn-Murray-Carvalho
 2020) applied to the differenced outcome ΔY = Y_post - Y_pre.

src/statspai/did/misclassified.py

@@ -1,6 +1,6 @@
 """
 Staggered Adoption DiD with Misclassification and Anticipation
-(Augustin, Gutknecht & Liu 2025, arXiv 2507.20415).
+(Augustin, Gutknecht & Liu 2025, arXiv 2507.20415). [@did_misclassified2025]
 
 Extends Callaway-Sant'Anna to handle two common practical
 imperfections: