PR #457 R1 polish: qualify docstring parity claim + refresh skip message

R1 verdict was Looks good with 2 P3 informational items. Both addressed:

1. P3 (Documentation/Tests): `bacon_decompose()` docstring example
   said "matches R bacondecomp::bacon() at atol=1e-6" without mentioning
   the documented always-treated convention exception. Qualified the
   example to spell out the aggregate-vs-per-component split:
   aggregate parity holds for all panels at atol=1e-6, per-component
   parity holds when first_treat is bounded below by min(time) (no
   always-treated), and the divergence on always-treated panels is by
   convention (Python remap-to-U vs R's `Later vs Always Treated`).
   Cross-references the REGISTRY note for the full contract.

2. P3 (Documentation/Tests): `TestBaconParityR`'s skip message still
   said the goldens were "deferred until R is provisioned (see TODO.md)"
   but the TODO row was removed in this PR. Updated to describe the
   intended skip case (partial-checkout / packaging scenarios where
   the committed JSON is unavailable) and dropped the TODO reference.

Tests unchanged: 33/33 pass in test_methodology_bacon.py.

Author: igerber

diff_diff/bacon.py

@@ -1302,8 +1302,15 @@ def bacon_decompose(
     >>> from diff_diff import bacon_decompose
     >>>
     >>> # Default: paper-faithful Goodman-Bacon (2021) Theorem 1 weights
-    >>> # (weights="exact"); matches R bacondecomp::bacon() at atol=1e-6
-    >>> # (validated via tests/test_methodology_bacon.py::TestBaconParityR).
+    >>> # (weights="exact"); matches R bacondecomp::bacon() at atol=1e-6 on
+    >>> # the aggregate (TWFE coefficient + weights-sum) across all panels,
+    >>> # and on the per-component breakdown when first_treat is bounded
+    >>> # below by min(time) (no always-treated). For panels with
+    >>> # always-treated units, the per-component breakdown diverges by
+    >>> # convention (Python remaps to U per paper footnote 11; R emits
+    >>> # `Later vs Always Treated`); see REGISTRY note on R parity
+    >>> # convention divergence. Validated via
+    >>> # tests/test_methodology_bacon.py::TestBaconParityR.
     >>> results = bacon_decompose(
     ...     data=panel_df,
     ...     outcome='earnings',

tests/test_methodology_bacon.py

@@ -301,11 +301,12 @@ def test_eq_9_later_vs_earlier_variance(self) -> None:
 def _load_r_golden() -> dict:
     if not _R_GOLDEN_PATH.exists():
         pytest.skip(
-            f"R parity goldens missing at {_R_GOLDEN_PATH}. To generate, "
+            f"R parity goldens missing at {_R_GOLDEN_PATH}. To regenerate, "
             "install R + `install.packages('bacondecomp')` + "
             "`install.packages('jsonlite')` then `cd benchmarks/R && "
-            "Rscript generate_bacon_golden.R`. The R goldens are deferred "
-            "until R is provisioned (see TODO.md)."
+            "Rscript generate_bacon_golden.R`. The goldens are committed "
+            "to the repo by default; this skip path covers partial-checkout "
+            "or packaging scenarios where the JSON file is unavailable."
         )
     return json.loads(_R_GOLDEN_PATH.read_text())