docs: add CBWSDID (Ustyuzhanin 2026) paper review

Add the in-repo scholarly paper review for arXiv:2604.02293v1,
"Covariate-Balanced Weighted Stacked Difference-in-Differences"
(Vadim Ustyuzhanin, HSE, 2026), as the Step-1 methodology fidelity
artifact (PR-A) for a prospective CBWSDID estimator.

CBWSDID is a design-based extension of weighted stacked DID for
conditionally (rather than unconditionally) parallel untreated trends:
a within-sub-experiment matching/balancing design stage produces
nonnegative control design weights b_sa that compose with the Wing,
Freedman & Hollingsworth (2024) corrective stacked weights into a
single weighted least-squares stacked estimator. It nests weighted
stacked DID at b_sa = 1 and extends to repeated 0->1 / 1->0 episodes
under a finite-memory assumption.

The review transcribes the absorbing-adoption core (sub-experiment
construction, Q_sa corrective weights, design weights b_sa, final
stacked weights W_sa, the pooled estimator, and the W_sa-weighted
two-way-FE event-study regression), Assumptions 1-4, the repeated-
treatment extension with Assumptions R1-R6, the Section 5 inference
(unit-clustered cluster-robust conditional on the estimated design
weights, plus cluster-bootstrap options and the Abadie-Imbens (2008)
nonsmooth-matching caveat), the simulation and the Trounstine (2020)
and Acemoglu et al. (2019) applications, and flags the saturated-FE
regression delta versus the library's means-based StackedDiD as a PR-B
design question. The paper has no numbered equations/theorems and no
algorithm boxes, so all references are pinned to section numbers.

Docs-only and self-contained: no REGISTRY.md / doc-deps.yaml /
references.rst / code wiring (deferred to PR-B), matching the
new-estimator paper-review precedent (#497).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: igerber