When replying to a review, we generate HTML intended to appear exactly

like a comment would appear when loading the page, but with indications

that it's a draft. This generally works well, but one noticeable issue

had to do with avatars.
If using the fallback avatar (or any custom avatar backend that used

HTML rather than images for display), the result would be a broken

image. This is because the HTML assumed the presence of image URLs for

avatars, rather than the empty strings that an HTML-only avatar would

return.
Since HTML-based avatar rendering is the path forward, we now provide a

rendered avatar in the base template as part of RB.UserSession, and

provide a convenient RB.UserSession.getAvatarHTML() as well. The reply

editor view uses this instead of URLs to do the rendering.

Tested a new reply with the HTML-only fallback avatar and with Gravatars.

In both cases, I got a proper avatar.
Unit tests pass.