CHARACTER SET

This is the 8.0 version of the fix.

There were two problems:

1) The function REGEXP_REPLACE() and REGEXP_SUBSTR() didn't
convert the character set if their respective results and
plainly returned whatever ICU produced (i.e. UTF-16 with big
or little endian depending on the CPU)

2) The engine was using an inconsistent and inefficient
interface, sometimes returning std::string& and sometimes
String*.

Fixed by 1) converting the result back to whatever
agg_arg_charsets_for_comparison() dictates, 2) Only returning
indexes for Substr(), so that we can convert when called for.

Change-Id: Icf705f75079dee580580676dba62aa142b764e27