Currently, we first finish parsing the document and then localize all its elements.

With the move of Localization to Rust (bug 1660391), we can become more sophisticated and start applying localization onto elements as soon as the context is ready.

This may yield performance improvements since we don't have to place the element in DOM, then wait till the tree is done, then revisit all localization elements and apply translations.

This performance optimization may or may not pay off because we will need to trigger formatMessage per element, rather than a single call to formatMessages with all keys to translate.

But on paper, it seems like that should parallelize more of work and in optimistic scenario where we don't need to lazily trigger fallbacks should improve performance.