Address PR #401 R2 review (1 P3 maintainability)

R2 verdict was "Looks good". One P3 housekeeping item addressed:

P3 (maintainability): selector-aware messaging on the empty-state
path:

1. `_compute_path_effects` empty-selection warning now branches on
   `paths_of_interest is not None`. Under `paths_of_interest`, the
   message reads "every user-specified path was either unobserved
   ... per-path 'zero observed groups' UserWarnings already issued"
   so the user gets a clear pointer to the per-path warnings rather
   than a generic "no observed treatment path" message that
   misdescribes the cause. Under `by_path=k`, the original message
   is preserved.

2. `summary()` empty-state header changed from
   "Treatment-Path Disaggregation (by_path)" to
   "Treatment-Path Disaggregation"; the trailing parenthetical now
   says "by_path / paths_of_interest was a no-op on this panel".

Adds regression `test_paths_of_interest_all_unobserved_emits_distinct_warning`
asserting the paths_of_interest-specific warning text fires when
every requested path is unobserved.

Also fixes 7 pre-existing F541 f-string-without-placeholder warnings
in the warning blocks via `ruff --fix`.

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

Author: igerber

diff_diff/chaisemartin_dhaultfoeuille.py

@@ -5807,16 +5807,34 @@ def _compute_path_effects(
     )
 
     if not selected_paths:
-        warnings.warn(
-            f"by_path / paths_of_interest was requested but no observed treatment "
-            f"path has a complete window [F_g-1, F_g-1+L_max={L_max}] "
-            f"within the panel. results.path_effects is populated as an "
-            f"empty dict to signal 'requested but empty'. Extend the "
-            f"panel so switchers have L_max+1 consecutive observed cells "
-            f"starting at F_g-1, or reduce L_max.",
-            UserWarning,
-            stacklevel=2,
-        )
+        if paths_of_interest is not None:
+            # Every requested path was unobserved (each emitted its own
+            # per-path "zero observed groups" warning inside the
+            # enumerator). Distinguish from the by_path=k case where
+            # the panel itself has no complete-window path.
+            warnings.warn(
+                "paths_of_interest was requested but every "
+                "user-specified path was either unobserved in the "
+                "panel or had a window outside the L_max+1 "
+                "convention (per-path 'zero observed groups' "
+                "UserWarnings already issued). results.path_effects "
+                "is populated as an empty dict to signal 'requested "
+                "but empty'.",
+                UserWarning,
+                stacklevel=2,
+            )
+        else:
+            warnings.warn(
+                f"by_path={by_path} was requested but no observed "
+                f"treatment path has a complete window [F_g-1, "
+                f"F_g-1+L_max={L_max}] within the panel. "
+                f"results.path_effects is populated as an empty dict "
+                f"to signal 'requested but empty'. Extend the panel "
+                f"so switchers have L_max+1 consecutive observed "
+                f"cells starting at F_g-1, or reduce L_max.",
+                UserWarning,
+                stacklevel=2,
+            )
         return {}
 
     # Cohort ids for the variance-eligible set (same construction as the

diff_diff/chaisemartin_dhaultfoeuille_results.py

@@ -1270,10 +1270,11 @@ def _render_path_effects_section(
             lines.extend(
                 [
                     thin,
-                    "Treatment-Path Disaggregation (by_path)".center(width),
+                    "Treatment-Path Disaggregation".center(width),
                     thin,
                     "  No observed paths have a complete [F_g-1, F_g-1+L_max] window.",
-                    "  (See UserWarning emitted at fit(); by_path was a no-op " "on this panel.)",
+                    "  (See UserWarning emitted at fit(); by_path / "
+                    "paths_of_interest was a no-op on this panel.)",
                     thin,
                     "",
                 ]

tests/test_chaisemartin_dhaultfoeuille.py

@@ -8732,6 +8732,37 @@ def test_paths_of_interest_frequency_rank_is_true_frequency(self):
         assert res.path_effects[(0, 1, 1, 1)]["frequency_rank"] == 1
         assert res.path_effects[(0, 1, 0, 0)]["frequency_rank"] == 2
 
+    def test_paths_of_interest_all_unobserved_emits_distinct_warning(self):
+        """When every path in `paths_of_interest` is unobserved,
+        the empty-state warning should mention `paths_of_interest`
+        explicitly rather than the generic `by_path={n}` text
+        (regression for R2 P3 maintainability finding)."""
+        df = _by_path_three_path_data()
+        est = ChaisemartinDHaultfoeuille(
+            drop_larger_lower=False,
+            paths_of_interest=[(1, 1, 1, 1), (1, 0, 1, 0)],  # both unobserved
+            twfe_diagnostic=False,
+            seed=42,
+        )
+        with warnings.catch_warnings(record=True) as recorded:
+            warnings.simplefilter("always", UserWarning)
+            res = est.fit(
+                df, outcome="outcome", group="group", time="period",
+                treatment="treatment", L_max=3,
+            )
+        # The summary empty-state warning mentions paths_of_interest, not by_path
+        empty_state_warnings = [
+            str(w.message)
+            for w in recorded
+            if "paths_of_interest was requested but every" in str(w.message)
+        ]
+        assert len(empty_state_warnings) >= 1, (
+            f"Expected paths_of_interest-specific empty-state warning, "
+            f"got: {[str(w.message) for w in recorded]}"
+        )
+        # And the result is empty dict, not None
+        assert res.path_effects == {}
+
     def test_unobserved_path_warns_and_omits(self):
         df = _by_path_three_path_data()
         est = ChaisemartinDHaultfoeuille(