Also fixed a bug where the GlobalKeymap wasn't returning true when it
matched a binding, which caused key events to be processed twice when
they bubbled out of the editor and hit the root view.
Ace triggers key handlers for both keydown (onCommandKey) and input
(onTextInput) events. Input events don't pass the event, which was
blowing up the keymap.
Now if a keypress event bubbles to an element with bindings for
multiple matching selectors, the most specific selector is chosen. This
allows us to override a keybinding by using a more specific selector.
For example, say we have an editor instance inside of the file-finder
widget, so that the user can use all their normal bindings when they
type the name of the file. Except when they hit <esc> we want to close
the file finder, not do whatever they normally have it bound to. Now we
can just say:
```
.file-finder .editor {
<esc>: close-file-finder
}
```
And we're assured that our binding will take precedence.
Say we have a structure like:
div.parent
div.child
div.grandchild
And we have two mappings:
.parent { x: foo }
.child { x:bar }
If there's an event originating on grandchild, it will *only* trigger
bar, because that selector selects a closer ancestor.
KeyEventHandler holds references to BindingSets. The name "binding set"
is based on the concept of a CSS ruleset. The idea is to choose a key
binding for an event based on what selectors (match / most closely
contain) the event's target DOM node.
If we have pattern 'da' and 'dad' both mapped, when 'da' is typed, the KeyMap waits for a given timeout for another character to be entered. If the time elapses, it goes ahead and executes the action for 'da'. But if a 'd' is subsequently entered, it executes 'dad'.
Rename FileSystemHelper-contentsOfDirectoryAtPath… to -listFilesAtPath to make it clear that we're only listing files, not subdirectories. This is a fairly special purpose method but it saves us from calling back into objective-c a ton of times to filter them in JS, and makes bringing up the file finder ~2x as fast.
Converting from JS-cocoa wrapper values for every element in the array appears to be very slow. Here we construct native JS datatypes and it's much faster. This is very messy and leaky and needs to be cleaned up if we want to keep it.
