feat: adapt actuation receiver to state's qi_session

Makes actuation tests pass. In main, the timing of the socket no longer contains the time to receive and send data, but only the processing time of the message handler.

ref: N25B-168

src/robot_interface/endpoints/actuation_receiver.py

@@ -28,20 +28,22 @@ class ActuationReceiver(ReceiverBase):
             logging.warn("Received message to speak, but it lacks data.")
             return
 
+        if not isinstance(text, (str, unicode)):
+            logging.warn("Received message to speak but it is not a string.")
+            return
+
         logging.debug("Received message to speak: {}".format(text))
 
         if not state.qi_session: return
+        # If state has a qi_session, we know that we can import qi
+        import qi  # Takes a while only the first time it's imported
 
         if not self._tts_service:
             self._tts_service = state.qi_session.service("ALTextToSpeech")
 
-        self._tts_service.say(text)
+        # Returns instantly. Messages received while speaking will be queued.
+        qi.async(self._tts_service.say, text)
 
     def handle_message(self, message):
-        if "endpoint" not in message:
-            return {"endpoint": "error", "data": "No endpoint provided."}
-
         if message["endpoint"] == "actuate/speech":
             self._handle_speech(message)
-
-        return {"endpoint": "error", "data": "The requested endpoint is not supported."}