There are many small things we can optimize in serializers. Reduce virtual calls, some string handling optimization, etc.

Comment on attachment 454851 [details] [diff] [review] wip2 So this moves mBodyOnly and mInBody from XHTMLSerializer to XMLSerializer so that few methods can be de-virtualized. In theory HTMLAtomTagToId -> HTMLCaseSensitiveAtomTagToId change does affect to behavior. But it should do now the right thing, since we get the atom always from an element, and HTML elements have lowercase Tag(). In fact, I don't know why we have case-insensitive HTMLAtomTagToId. Other changes are addref/release removal related to SetCachedEncoder and faster sibling iteration.