- Implement the localization service.
- Use the proper build process which generates the require JSON files.
- Implement getting the locale and language configuration.

- Don't use "any" for the API type.
- Remove everything from the Coder API that can eventually be done
  through the VS Code API.
- Move the event emission to our own client to minimize patching.

Since this will be a path in the binary that we don't want to create on
the user's system. I also removed the option to override it; it doesn't
seem like a great idea since you'd always want those builtin extensions.
This way we also don't have to check if the option was passed and only
create it if that was the case.

This isn't a real error event; we artificially emit it just in case
something waiting to start is listening to the error event in order to
clean up and/or restart.

* Allow setting paths for builtin exts and extra dirs

The extra directories aren't used yet, just available from the
environment service and to the shared process.

* Utilize extra builtin extensions path

* Utilize extra extensions directory

* Fix cached mtimes for extra extension dirs

* Simplify extension cache equality check

* Fix hash

* Remove whitespace

Mac was using its own thing.

- Fixes #535
- Fixes #501

This normally happens in electron-main/main.ts but we don't run anything
from there.

Fixes #499.

Since this is where it attempts to read it, this should solve the JSON
errors for good.

* Update VS Code to 1.33.0

* Fix slow file tree

* Fix WindowsService fill

* Provide `off` on event listeners

* Fix webview

* Fix double title bar and missing preferences on Mac

* Bump VS Code version in Travis config

* Fix black dialog text (again)

* Fix shared process not starting

Registering returns an instance that lets you retry and recover without
needing to keep passing the name everywhere.

Also refactored the shared process a little to make better use of the
retry and downgraded stderr messages to warnings because they aren't
critical.

* Make proxies decide how to handle disconnects

* Connect to a new terminal instance on disconnect

* Use our retry for the watcher

* Specify method when proxy doesn't exist

* Don't error when closing/killing disconnected proxy

* Specify proxy ID when a method doesn't exist

* Use our retry for the searcher

Also dispose some things for the watcher because it doesn't seem that
was done properly.

The searcher also now starts immediately so there won't be lag when you
perform your first search.

* Use our retry for the extension host

* Emit error in parent proxy class

Reduces duplicate code. Not all items are "supposed" to have an error
event according to the original implementation we are filling, but there
is no reason why we can't emit our own events (and are already doing so
for the "disconnected" event anyway).

* Reconnect spdlog

* Add error message when shared process disconnects

* Pass method resolve to parse

* Don't pass method to getProxy

It doesn't tell you anything that trace logging wouldn't and has
no relation to what the function actually does.

* Fix infinite recursion when disposing protocol client in tests

* Replace evaluations with proxies and messages

* Return proxies synchronously

Otherwise events can be lost.

* Ensure events cannot be missed

* Refactor remaining fills

* Use more up-to-date version of util

For callbackify.

* Wait for dispose to come back before removing

This prevents issues with the "done" event not always being the last
event fired. For example a socket might close and then end, but only
if the caller called end.

* Remove old node-pty tests

* Fix emitting events twice on duplex streams

* Preserve environment when spawning processes

* Throw a better error if the proxy doesn't exist

* Remove rimraf dependency from ide

* Update net.Server.listening

* Use exit event instead of killed

Doesn't look like killed is even a thing.

* Add response timeout to server

* Fix trash

* Require node-pty & spdlog after they get unpackaged

This fixes an error when running in the binary.

* Fix errors in down emitter preventing reconnecting

* Fix disposing proxies when nothing listens to "error" event

* Refactor event tests to use jest.fn()

* Reject proxy call when disconnected

Otherwise it'll wait for the timeout which is a waste of time since we
already know the connection is dead.

* Use nbin for binary packaging

* Remove additional module requires

* Attempt to remove require for local bootstrap-fork

* Externalize fsevents

There is a window between when the stat is made and the result makes it
back to the client where a file is created or deleted and it won't be a
part of that stat.

To fix it, I added a new property signaling that we got some changes
since starting the stat and should run the stat again because we have no
way of knowing whether the currently running stat will include the new
changes or not.

Forgot to move this to the new file when upgrading to 1.32.0.

* Update VS Code to 1.32.0

* Update patch

Most changes are moved files, most notably shell.contribution.ts which
is now main.contribution.ts.

Also:
- repl.ts no longer uses isMacintosh
- shell.ts doesn't exist
- added back the commented-out CSP headers

* Use es6 target for bootstrap-fork

* Directly reference cross-env binary

yarn and npm find the binary just fine when running the tasks from the
root but it doesn't work if you run one of those tasks directly from
within those directories.

* Update import paths and bootstrap-fork ignores

* Increase memory limit for building default extensions

* Fix invalid regex in Firefox

* Update startup function

* Fix global.require error

* Update zip extract arguments

* Update travis to minimum required Node version

* Always chmod executable dependencies

Fixes EACCESS errors for users that had the files unpacked before we
added the chmod call.

* Remove unused var declaration