fix: align dtypes between context and future DataFrames for Chronos-2

Chronos-2 predict_df() validation requires matching dtypes for columns appearing in both context and future DataFrames. After masking (setting NaN for D+2-D+14 in partial_d1 features), pandas converts int64 columns to float64. This caused:

ValueError: Column lta_total_allocated in future_df has dtype float64 but column in df has dtype int64

Solution: After masking, align dtypes by casting context columns to match future column dtypes.

Co-Authored-By: Claude <[email protected]>

src/forecasting/dynamic_forecast.py

@@ -83,6 +83,18 @@ class DynamicForecast:
         # Step 3: Apply availability masking
         future_data = self._apply_masking(future_data, run_date)
 
+        # Step 4: Align dtypes between context and future
+        # Chronos-2 requires matching dtypes for columns that appear in both DataFrames
+        # After masking, int columns may become float due to NaN values
+        common_cols = set(context_data.columns) & set(future_data.columns)
+        for col in common_cols:
+            if col in ['timestamp', 'border']:
+                continue  # Skip non-numeric columns
+            # If context has int and future has float (due to NaN), cast context to float
+            if context_data[col].dtype != future_data[col].dtype:
+                # Use the dtype from future_data (which may be float64 due to NaN masking)
+                context_data[col] = context_data[col].astype(future_data[col].dtype)
+
         return context_data, future_data
 
     def _extract_context(