Address PR #371 AI review R3: docstring access-pattern accuracy

P3 fix: `_collect_path_placebo_bootstrap_inputs` docstring described
the analytical-results shape as
`path_placebos[path]["horizons"][-lag_l]["effect"]` but the actual
access is `path_placebos[path][-lag_l]["effect"]` (no `["horizons"]`
wrapper -- `_compute_path_placebos` returns the negative-keyed inner
dict directly, intentionally diverging from `_compute_path_effects`'s
`["horizons"]` nesting). Update the docstring to match the actual
access pattern at the implementation site (`:5818-5824`).

Harmless at runtime; the fix is to prevent the comment from misleading
future maintenance.

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

Author: igerber

diff_diff/chaisemartin_dhaultfoeuille.py

@@ -5734,9 +5734,12 @@ def _collect_path_placebo_bootstrap_inputs(
     ``path_placebo_event_study[path][-lag_l]`` post-bootstrap.
 
     The point estimate per ``(path, lag_l)`` is read from
-    ``path_placebos[path]["horizons"][-lag_l]["effect"]`` to stay
-    bit-identical with the analytical pass; the bootstrap distribution
-    gets centered on this value by ``_bootstrap_one_target`` downstream.
+    ``path_placebos[path][-lag_l]["effect"]`` (note: no ``["horizons"]``
+    wrapper -- ``_compute_path_placebos`` returns the negative-keyed
+    inner dict directly, unlike ``_compute_path_effects`` which wraps
+    its horizons under a ``["horizons"]`` key) to stay bit-identical
+    with the analytical pass; the bootstrap distribution gets centered
+    on this value by ``_bootstrap_one_target`` downstream.
 
     The ``warnings.catch_warnings`` block suppresses the
     re-enumeration overflow ``UserWarning``; the analytical