Remove the broken optimization from UnicodeToANSI() that attempted

Remove the broken optimization from UnicodeToANSI() that attempted
in a stupid way to avoid to WideCharToMultiByte() call by testing
to see if the second byte (not wchar) was NUL.  This test works
for Latin1 but not for anything more interesting.  Always call
WideCharToMultiByte().

I commented out the use of WC_NO_BEST_FIT_CHARS in order to
provide compatibility with Windows.  Windows converts the user
name without that option.

With the current code a principal name consisting of a single
component equivalent to the Greek character Sigma and a realm
name will be converted to S@REALM exactly as the "WhoAmI"
command does.  If WC_NO_BEST_FIT_CHARS was specified, this
string would be converted to "?@REALM".

ticket: 5766

git-svn-id: svn://anonsvn.mit.edu/krb5/trunk@19970 dc483132-0cff-0310-8789-dd5450dbe970