The changes here stop winit from automatically appending the window canvas to the document root in the web platform. I discussed this with @ryanisaacg on Discord and we seemed to agree.

I think this approach has multiple benefits:

• It reduces the scope of winit to simply providing a canvas with the proper event listeners attached.
• We get multiple canvas support for free, given that all the event listeners are currently local to the canvas (however, we still need to think if we should provide fullscreen support and how in Web support tracking issue #1072).
• The user decides how to embed the canvas freely in their own web document. This allows us to leave some functionality out of winit's scope, like changing the document/tab title or updating the favicon. The user is expected to implement this on top of winit if necessary.

Additionally, I have updated an unnecessary TODO comment and fixed some panics on canvas creation.