Make cmp() work on EBCDIC with both UTF-8 operands

Make cmp() work on EBCDIC with both UTF-8 operands

Prior to this commit, comparing two UTF-EBCDIC encoded strings did not
work properly, due to the complexities of UTF-EBCDIC in the code point
0-255 range.  This commit adds specialized code to handle this.  The
code could be enabled and work properly on ASCII platforms, but isn't
because a simple memcmp() suffices for these.  Since sort() uses cmp(),
this also fixes UTF-EBCDIC sort.