Add EDID validation tests against Chen, Sant'Anna & Xie (2025)

Validate the EfficientDiD estimator against published results from the
paper's Table 6 (HRS empirical application) and Tables 4/5 patterns
(Compustat MC simulations).

HRS replication: point estimates match within 0.1-1.3% for all targets
except ATT(9,10) which is a near-zero noisy estimate (90 ± 641). The
CS cross-validation matches to <$1 on all ATT(g,t), confirming data
loading correctness. Fixture is a 656-individual subset of the Dobkin
et al. (2018) publicly available replication kit.

MC simulations validate unbiasedness, RMSE dominance over CS, efficiency
gains increasing with serial correlation, ~95% coverage, and SE
calibration across rho values.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: igerber

.gitignore

@@ -90,3 +90,6 @@ papers/
 
 # Local analysis notebooks (not committed)
 analysis/
+
+# Replication data (local only, not for distribution)
+replication_data/