This eliminates quite a lot of UTF-8 / UTF-16 churn.
This is required for upcoming change where we want to emit ThrowIfTDZ for assignment expressions only for lexical declarations.