fix: use renamed 'target' column instead of 'target_border_X' in inference

The dynamic_forecast.prepare_forecast_data() method renames the target column
from 'target_border_X' to just 'target', but the inference code was still trying
to access the original column name, causing KeyError.

This was the root cause of empty forecasts - the model never received target data.

src/forecasting/chronos_inference.py

@@ -169,8 +169,8 @@ class ChronosInferencePipeline:
                     border=border
                 )
 
-                # Get target column name
-                target_col = f"target_border_{border}"
+                # Get target column name (note: dynamic_forecast renames it to 'target')
+                target_col = 'target'
 
                 # Extract context values
                 context = context_data[target_col].values