PI: Fix broadcast of timeseries over all ensembles

When broadcasting a timeseries to all ensemble members, we have to be
sure to do so _after_ it has been filled (with NaNs) to the right
length.

src/rtctools/data/pi.py

@@ -575,13 +575,6 @@ def __init__(self, data_config, folder, basename, binary=True, pi_validate_times
                 unit = header.find('pi:units', ns).text
                 self.set_unit(variable, unit=unit, ensemble_member=ensemble_member)
 
-                if make_virtual_ensemble:
-                    # Make references to the original input series for the virtual
-                    # ensemble members.
-                    for i in range(1, self.ensemble_size):
-                        self.__values[i][variable] = self.__values[0][variable]
-                        self.set_unit(variable, unit=unit, ensemble_member=i)
-
                 # Prepend empty space, if start_datetime > self.__start_datetime.
                 if start_datetime > self.__start_datetime:
                     if self.__dt:
@@ -614,6 +607,13 @@ def __init__(self, data_config, folder, basename, binary=True, pi_validate_times
                     self.__values[ensemble_member][variable] = np.hstack(
                         (self.__values[ensemble_member][variable], filler))
 
+                if make_virtual_ensemble:
+                    # Make references to the original input series for the virtual
+                    # ensemble members.
+                    for i in range(1, self.ensemble_size):
+                        self.__values[i][variable] = self.__values[0][variable]
+                        self.set_unit(variable, unit=unit, ensemble_member=i)
+
             if not self.__dt:
                 # Remove time values outside the start/end datetimes.
                 # Only needed for non-equidistant, because we can't build the