Add extensible "flags" to the kernel

[rupertsworld]

Right now, we have the concept of "display" modes. Each action has a display property, and if that property is missing or auto the model can choose its own display mode (e.g. snippet or inline) to control how the action's output displays.

However, different applications might have different display modes – for example, voice or watchface. And developers might want to extend the kernel with their own "flags" that describe other types of choices they want the model to make in its action decisions.

This issue proposes we extend the kernel with the idea of "flags" that can be added to the Interpreter. This also fits our ethos of allowing extensibility & control over any configuration and prompts (right now the display mode prompts are not possible to alter).

For example:

const flags = {};

flags.display = {
  description: 'The display mode for the action.'
  options: {
    voice: `The action is spoken to the user (useful when on handsfree mode)`,
    hud: `The action is shown in the user's heads-up display`,
  }
};

const interpreter = new Interpreter({ ..., flags });

Now, when we get an action response, these flags are present in the response.

const response = interpreter.generateActionResponse({ ... });
console.log(response.display) // 'voice'

Questions:

• Should we put response flags in response.[flagname] or response.flags.[flagname]?
• Like display modes, should we assume that actions can be defined in resources to set their own flags?
• Can/should flags also appear in messages?
• Does the ProcessRuntime need to know about them?