proposal for optimization in Unicode comparisons

[Sorin Marian Nasoi]

I had a look today into the ICU documentation and I found something that I think we can use to optimize our usage of ICU:

Starting with ICU 4.2 they introduced
http://icu-project.org/apiref/icu4c/classStringPiece.html
that handle UTF-8 strings comparisons

This means that we do not need to do conversions from
"const char*" (our UTF-8 encoded strings) to
"UnicodeString" (ICU's internal UTF-16 strings)
for comparing 2 strings using ICU.

So, instead of using:
Collator::compare(const UnicodeString & , const UnicodeString & )

we can use:

Collator::compareUTF8( const StringPiece & , cnst StringPiece & , UErrorCode & )

What I propose is this:
- if the user has linked against ICU version greater or equal to 4.2, use the "StringPiece" class
- if the ICU version is less then 4.2, keep current behaviour for example, in src/util/utf8_util.h starting at line 682, instead of:

 unicode::string us1;
 unicode::string us2;

 unicode::to_string(s1, &us1);
 unicode::to_string(s2, &us2);

 Collator::EComparisonResult result = ::Collator::EQUAL;

 result = static_cast<Collator*>(collation->getCollator())->compare(us1, us2);

 return result;

do

 ::UCollationResult result = ::UCOL_EQUAL ;
 UErrorCode status = U_ZERO_ERROR;

 result = static_cast<Collator*>(collation->getCollator())->compareUTF8(s1.c_str(),
s2.c_str(), status);

 if(U_FAILURE(status))
 {
   assert(false);
 }

 return result;