Move locate to DOM Renderers

[Zynton]

We need to move this locate stuff. It is DOM stuff so it should be on the renderer. If it was ideed on the renderer, you could simply read the engine.renderings object rather than having to "observe" your DOM surroundings. Something like this:

class FontAwesomeDomRenderer {
    locate(domNode: Node, offset: number, node: VNode): [this, RelativePosition] {
        const rendering = this.engine.renderings.get(node);
        if (domNode === rendering[rendering.length - 1]) {
            // We are targetting the invisible character AFTER the
                // fontawesome node.
            if (offset === 0) {
                // Moving from after the fontawesome node to before it.
                // (DOM is `<invisible/><fontawesome/>[]<invisible/>` but
                // should be `[]<invisible/><fontawesome/><invisible/>`).
                return [this, RelativePosition.BEFORE];
            } else {
                // Stay after the fontawesome node.
                return [this, RelativePosition.AFTER];
            }
        } else if (domNode === rendering[0]) {
            // We are targetting the invisible character BEFORE the
            // fontawesome node.
            if (offset === 1) {
                 // Moving from before the fontawesome node to after it.
                // (DOM is `<invisible/>[]<fontawesome/><invisible/>` but
                // should be `<invisible/><fontawesome/><invisible/>[]`).
                return [this, RelativePosition.AFTER];
            } else {
                // Stay before the fontawesome node.
                return [this, RelativePosition.BEFORE];
            }
        }  else {
            // We should never reach this. It's a fallback to prevent a crash in
            // the eventuality that it would.
            return [this, RelativePosition.BEFORE];
        }
    }
}

In particular, your code will fail in the event that some node that is not a FontAwesomeNode ends up rendering itself using a node that your regex would recognize as being a fontawesome. If it does, the first invisible character would be mistaken for the second one because its "previousSibling" would be recognized as a fontawesome even though it is not the one that this invisible character is "attached" to. Hence my suggestion of using the renderings object to specifically work on the nodes rendered by that VNode and only that.