Doesn't the library just have to perform a simple search-and-replace on the entire page's HTML for the blacklisted characters to remove them?

Normally blacklisting is bad, but we're only targeting the few text-direction-changing characters that exist.