Parse provided holiday list into a dataframe

Summary:
Currently, we pass `self.holidays` into `get_holiday_dates` to infer all the holidays that require adjustments to anomaly scores.

Unfortunately, `self.holidays` is `None` in predict calls -- it's only [instantiated](https://fburl.com/code/pauwfwhb) when calling `fit()`.

In 1D, since we create a new ProphetModel instance when calling `predict()` (with the underlying fbprophet model passed in), `self.holidays` will be None on that recreated instance.

Thus, explicitly create a dataframe with the holidays on predict calls.

Reviewed By: irumata

Differential Revision: D56586165

fbshipit-source-id: df156e7bd6d5ecbbd47674d769f8316788c98c02

kats/detectors/prophet_detector.py

@@ -590,8 +590,12 @@ def predict(
             self.holiday_multiplier is not None
             and round(self.holiday_multiplier, 10) != 1.0
         ):
+            # convert the list of custom holidays into a df
+            custom_holidays = (
+                pd.DataFrame(self.holidays_list) if self.holidays_list else None
+            )
             holidays_df: Optional[pd.Series] = get_holiday_dates(
-                self.holidays, self.country_holidays, data.time
+                custom_holidays, self.country_holidays, data.time
             )
             if holidays_df is not None:
                 scores_ts = pd.Series(list(scores.value), index=data.time)