The intl extension supports the grapheme concept from Unicode,
while the mbstring extension handle UT8 codepoints.

An emoji like 🏴󠁧󠁢󠁥󠁮󠁧󠁿 has 28 bytes, 7 codepoints (tags E N G L A N D),
1 grapheme. That could affects method like substr or strlen when
we want to manipulate graphemes and not codepoints.

Strategy is to offer bytes/codepoints/graphemes capabilities,
downgrade from graphemes to codepoints for non UTF-8 encoding,
and defaults to grapheme.