Fix scheduled update time-drift in data update coordinator (#32974)

* Fix scheduled update time-drift in data update coordinator

As next schedule is calculated **after** the update is done,
setting now + update_interval makes 1 second drift in practice,
as the tick is 1Hz.

* Floor the utcnow timestamp

to remove sub-second error precision, and generate constant frequency updates
as long as the update takes < 1s.

homeassistant/helpers/update_coordinator.py

@@ -88,8 +88,14 @@ class DataUpdateCoordinator:
             self._unsub_refresh()
             self._unsub_refresh = None
 
+        # We _floor_ utcnow to create a schedule on a rounded second,
+        # minimizing the time between the point and the real activation.
+        # That way we obtain a constant update frequency,
+        # as long as the update process takes less than a second
         self._unsub_refresh = async_track_point_in_utc_time(
-            self.hass, self._handle_refresh_interval, utcnow() + self.update_interval
+            self.hass,
+            self._handle_refresh_interval,
+            utcnow().replace(microsecond=0) + self.update_interval,
         )
 
     async def _handle_refresh_interval(self, _now: datetime) -> None: