Rewrite RC reg/DR to match DRDID::reg_did_rc and DRDID::drdid_rc formulas

_outcome_regression_rc: Pool all treated obs for OR correction term
(was: separate per-period averages). Period-specific treated means for Y.
Matches Sant'Anna & Zhao (2020) Eq 2.2 / R reg_did_rc exactly.

_doubly_robust_rc: Fit 4 OLS models (control+treated, pre+post) for
locally efficient DR estimator (was: 2 control-only). Implements tau_1
(AIPW) + tau_2 (local efficiency adjustment) with full 11-component IF.
Matches Sant'Anna & Zhao (2020) Eq 3.3+3.4 / R drdid_rc exactly.

Add agg_weight field to group_time_effects for RCS aggregation weight
(cohort mass), separate from n_treated (per-cell display count).
Aggregation uses data.get("agg_weight", data["n_treated"]) for
backward compatibility with panel data.

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

Author: igerber