Add support for continuous speech recognition

We currently report RECOGNITION_COMPLETE upon reception of the first result.
We should permit the client to specify continuous recognition so that it can be used for example with tests with a microphone.