Fix coordinator wiring after refactor

Two bugs left from the hass.data move that surfaced as soon as the
integration ran on the live HA:

* In async_setup_entry, devices were stored in hass.data AFTER
  coordinator.async_config_entry_first_refresh(), but
  _async_update_data reads them from hass.data — so the very first
  refresh raised KeyError 'devices' and the entry stayed in
  setup_retry. Move the hass.data update before first_refresh.

* SmartThingsDeviceTracker called self.async_on_update(...) which
  isn't a method on TrackerEntity. Without a working listener
  registration the tracker stayed restored=true / unavailable
  forever. Subscribe in async_added_to_hass and use async_on_remove
  for cleanup, which is the documented pattern.

Verified end-to-end on a live HA: phones/watch produce GPS, battery
sensors populate, and the Ring button triggers a SmartTag.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

custom_components/smartthings_find/device_tracker.py

@@ -43,8 +43,14 @@ class SmartThingsDeviceTracker(DeviceTrackerEntity):
 
         if 'icons' in device['data'] and 'coloredIcon' in device['data']['icons']:
             self._attr_entity_picture = device['data']['icons']['coloredIcon']
-        self.async_on_update(coordinator.async_add_listener(self.async_write_ha_state))
-    
+
+    async def async_added_to_hass(self) -> None:
+        """Subscribe to coordinator updates once added to HA."""
+        await super().async_added_to_hass()
+        self.async_on_remove(
+            self.coordinator.async_add_listener(self.async_write_ha_state)
+        )
+
     def async_write_ha_state(self):
         if not self.enabled:
             _LOGGER.debug(f"Ignoring state write request for disabled entity '{self.entity_id}'")