This matches better with the other routes.

Also add a missing authentication check to the path proxy web socket.

This is mostly so we don't have to do any wacky patching but it also
makes it so we don't have to keep checking if the request is a web
socket request every time we add middleware.

This is for applications like Jupyter that aren't base path agnostic.

A provider can now be registered on multiple endpoints (or potentially
moved if needed).

This makes the code much more internally consistent (providers just
return payloads, include the proxy provider).

There's no need to specially construct the path.

Otherwise they'll crash code-server.

This simplifies the logic a bit.

The cookie will be set for the proxy domain so it'll work for all of its
subdomains.

It'll be able to handle /proxy requests as well as subdomains.