Implement streaming instead of request/response cycle

[cmgriffing]

The application currently waits for the entire audio clip to be recorded and then it passes it to Whisper. On a fast enough machine the audio should be quicker to process than its runtime, so we should implement a streamed version of the processing rather than waiting for the whole clip.

This will require a lot of thought around how the UI should present things while recording. We need to also consider how to cancel and stop recording properly.