lib-fts: Fix address tokenizer to handle large input properly

lib-fts: Fix address tokenizer to handle large input properly

Previously it could have used excessive amounts of memory if the input
didn't contain separator characters.

The fix changes a bit how the address-tokenizer works: Previously large
email addresses were saved as truncated tokens. Now they're skipped
entirely by the address tokenizer. Similarly when searching long email
addresses they're no longer searched as truncated tokens, but instead
simply fed to the parent tokenizer which (likely) searches them in
smaller pieces.

Note that this also sometimes changes the order in which tokens are
returned, e.g. "foo", "example", "foo@example.com", "com" instead of
returning "com" before the email address. This isn't ideal, but fixing it
seems annoyingly complicated and practically it doesn't matter right now.