Description:
------------
"Only the first character of thousands_sep is used. For example, if you use
& as thousands_sep on the number 1000, number_format() will return
1&000."

This example is based on US-ASCII or other one-byte character sets. If for
example the *single character* "ü" in UTF-8 is used, the byte-representation is
two byte (I might be wrong, but this is not relevant). The function
number_format() uses the *first byte* of that character, resulting in something
broken, like broken UTF-8.

Expected result:
----------------
Add the following note:
Multi-byte characters are *not*  supported in this function, as only the first byte
will be used, resulting in broken encoding.

Furthermore:
Only "thousands_sep" has this notice as of now, but according to my tests, the
same applies to "dec_point". So please add the note above there as well.

Actual result:
--------------
> php -r "var_dump(number_format(123456789.123, 4, 'ö', 'ä'));"
string(16) "123�456�789�1230"