Related to #31992.

This accepts user names that contain a single script, but not mixed-script
names, such as ones that mix Latin and Cyrillic. That seemed to be closest
to the code's existing philosophy.

Since PHP 7.2 and 7.3 don't offer mb_str_split, this change leaves sites
on those versions of PHP with ASCII-only user names.
 - Existing user names continue to work unchanged even on 7.2/7.3
 - adding new non-ASCII user names requires 7.4
 - autocorrection requires 7.4 (e.g. user'@example.com -> user@example.com)

Until now, WordPress generated all-ASCII slugs, which works well for
mostly-ASCII languages. Now that WordPress supports scripts like
Devanagari much better, it's necessary to test the slug support for
Devanagari. What WordPress did until now works well for Devanagari, so
this commit merely adds a unit test.

The ideal is to have links such as
	<hebrewdomain>.<hebrewtld>/<hebrewpath>
which is shown onscreen as
	<hebrewpath>/<hebrewtld>.<hebrewdomain>
without any switches between right-to-left Hebrew and left-to-right Latin.
/author/ and a few others don't achieve that ideal. Yet.

Note that there is a rare existing bug: The existing mapping to ASCII can
produce the same slug for two pages, since two letters can map to e.g.
'L'. This seems unfixable, since any change might break existing links,
but happily also extremely rare.

Suggested by Mukesh Panchal.

I also added some tests for sanitize_email, for which there were none
before. They didn't break, I just felt that having some is better than
having none.

No behaviour changes are needed due to these tests, but the test code
should capture the reasons for rejecting some of these addresses, so this
commit adds tests and comments.

Props tusharbharti.