POC of cursorless non-text API

[pokey]

Why?

We want to apply cursorless grammar / decorations to non-textual entities such as a file browser, email client, cursorless cheatsheet interface, etc

How?

We'll do a POC of a cursorless file browser. We proceed in the following milestones:

Milestone 1

1. Build a third-party extension that renders a file tree for a particular directory in a webview
2. The extension calls a cursorless api function, passing a list of tokens to be decorated
3. Cursorless does the following:
   1. adds these tokens to its list of all editor tokens, adding some field pointing back to the token given by third-party
   2. returns a list of decoration info back to third-party including svgs and character offsets so that extension can do rendering
4. Extension renders the hats in the appropriate place in its webview

Milestone 2

• Relies on Switch targets to object-oriented #210

5. When cursorless receives a command from talon, it can detect that the mark was from third party and call function on third-party to get the target. It should probably leverage new pipelines at this point, so we'll want to hold off on this impl for now

Milestone 3

• Relies on state sharing (eg Expose shape enablement #287)

6. When cursorless processes a scope type that is third-party, it will call a function on third-party to process modifier

Questions

• How do we handle focus? The webview may or may not be focused
• Somehow, extension should be able to register new scope types both extension and talon-side. Probably want state sharing (like Expose shape enablement #287) for this one

Philosophical points / future directions

We view text files as a special case, and can refactor our language / text support to rely on the same general purpose API