Address CI review: default n_simulation_failures=0 + all-failed test

Two items from CI AI review on PR #326:

1. P2 backward-compat: moving `n_simulation_failures` to the end of
   `SimulationPowerResults` with a default of `0`. Users who manually
   instantiate the dataclass with the pre-PR field order continue to
   work; `simulate_power()` still fills the field in via keyword. The
   field remains part of `to_dict()` output (PR-level contract
   unchanged).

2. P3 coverage: adding a regression test for the all-failed escape
   path. An estimator that raises `ValueError` on every replicate now
   asserts both the `RuntimeError("All simulations failed. ...")`
   message and that the narrow-except filter doesn't swallow it.

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

Author: igerber

diff_diff/power.py

@@ -1026,7 +1026,6 @@ class SimulationPowerResults:
     mean_se: float
     coverage: float
     n_simulations: int
-    n_simulation_failures: int
     effect_sizes: List[float]
     powers: List[float]
     true_effect: float
@@ -1039,6 +1038,7 @@ class SimulationPowerResults:
     survey_config: Optional[Any] = field(default=None, repr=False)
     mean_deff: Optional[float] = None
     mean_icc_realized: Optional[float] = None
+    n_simulation_failures: int = 0
 
     def __post_init__(self):
         """Compute derived statistics."""

tests/test_power.py

@@ -626,6 +626,24 @@ def fit(self, data, **kwargs):
                 data_generator=generate_did_data,
             )
 
+    def test_simulation_all_failed_raises_runtime_error(self):
+        """All simulations failing: narrow-except path still raises RuntimeError."""
+        from diff_diff.prep import generate_did_data
+
+        class AlwaysFailingEstimator:
+            def fit(self, data, **kwargs):
+                raise ValueError("every replicate fails")
+
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore", UserWarning)
+            with pytest.raises(RuntimeError, match="All simulations failed"):
+                simulate_power(
+                    estimator=AlwaysFailingEstimator(),
+                    n_simulations=5,
+                    progress=False,
+                    data_generator=generate_did_data,
+                )
+
     def test_simulation_failure_counter_survives_serialization(self):
         """`n_simulation_failures` round-trips through to_dict/to_dataframe."""
         from diff_diff.prep import generate_did_data