Fix deleting of stale marker for device lists (#6819)

We were in fact only deleting stale marker when we got an incremental
update, rather than when we did a full resync.

changelog.d/6819.misc

@@ -0,0 +1 @@
+Detect unknown remote devices and mark cache as stale.

synapse/storage/data_stores/main/devices.py

@@ -940,13 +940,6 @@ def _update_remote_device_list_cache_entry_txn(
             lock=False,
         )
 
-        # If we're replacing the remote user's device list cache presumably
-        # we've done a full resync, so we remove the entry that says we need
-        # to resync
-        self.db.simple_delete_txn(
-            txn, table="device_lists_remote_resync", keyvalues={"user_id": user_id},
-        )
-
     def update_remote_device_list_cache(self, user_id, devices, stream_id):
         """Replace the entire cache of the remote user's devices.
 
@@ -1003,6 +996,13 @@ def _update_remote_device_list_cache_txn(self, txn, user_id, devices, stream_id)
             lock=False,
         )
 
+        # If we're replacing the remote user's device list cache presumably
+        # we've done a full resync, so we remove the entry that says we need
+        # to resync
+        self.db.simple_delete_txn(
+            txn, table="device_lists_remote_resync", keyvalues={"user_id": user_id},
+        )
+
     @defer.inlineCallbacks
     def add_device_change_to_streams(self, user_id, device_ids, hosts):
         """Persist that a user's devices have been updated, and which hosts