Better support for auditory scanning

[willwade]

There are several key things you need for purely auditory scanning (ie accessing grid blind with a switch).

1. Read message bar as you write which means hearing the letters not just full words (see Add 'Speak label letter by letter' grid action #565)
2. Next letter prediction as an option. It just makes more sense when linear scanning.
3. Hearing predictions as you write. As you are entering your text if predictions are going to be useful then they need to be read aloud rapidly as you write. Somehow
4. A way of hearing the audio cue voice quietly (eg in a headphone) separate to the main voice (main speaker)

There's more to this but I'm picking the key aspects that asterics will struggle with. Eg splitting audio? The only way around that is making an app of the web app because, once again safari doesn't allow you to do this

// 1) Get permission so device labels are available
await navigator.mediaDevices.getUserMedia({ audio: true });

// 2) List output-capable devices
const devices = (await navigator.mediaDevices.enumerateDevices())
  .filter(d => d.kind === 'audiooutput');

// 3) Route an <audio> element to the chosen device
const audioEl = document.querySelector('audio'); // your player
await audioEl.setSinkId(devices[0].deviceId); // pick a deviceId