Preserve df=0 sentinel on DoseResponseCurve for NaN serialization

Pass _survey_df directly (including the 0 sentinel) to DoseResponseCurve
so that to_dataframe() also produces NaN inference for rank-deficient
replicate designs. Previously converted 0→None which caused
to_dataframe() to fall back to z-distribution with finite t_stat/p_value.

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

Author: igerber

diff_diff/continuous_did.py

@@ -684,7 +684,7 @@ def fit(
             target="att",
             p_value=att_d_p,
             n_bootstrap=self.n_bootstrap,
-            df_survey=_survey_df if _survey_df != 0 else None,
+            df_survey=_survey_df,
         )
         dose_response_acrt = DoseResponseCurve(
             dose_grid=dvals,
@@ -695,7 +695,7 @@ def fit(
             target="acrt",
             p_value=acrt_d_p,
             n_bootstrap=self.n_bootstrap,
-            df_survey=_survey_df if _survey_df != 0 else None,
+            df_survey=_survey_df,
         )
 
         # Strip bootstrap internals from gt_results