Issue1189

When using pythons md5 library to create a hex digest of a string
containing non-latin characters, an incorrect hash is returned. The
following code shows the difference:

# encoding: uft-8

from java.math import BigInteger
from java.security import MessageDigest
from java.security import NoSuchAlgorithmException
from java.lang import String
import md5

a = u"Gráin amháiñ ©ðƒ©óíßðƒóíıßðƒ‚íó©ı"
#a = "A lovely string to encrypt!"
b = String(a)

digest = MessageDigest.getInstance("MD5")
bytes = b.getBytes()
digest.reset()
java_hash = BigInteger(1, digest.digest(bytes)).toString(16)
print "Hash using Java:\t", java_hash

print "Hash in Python:\t\t", md5.new(a).hexdigest()

CPython 2.5 takes str (byte strings), automatically coercing unicode
string via a encode to ascii as necessary.

So this is not even legal in CPython, since it throws
UnicodeEncodeError: 'ascii' codec... Nonetheless, we need more
investigation.

To correct some ambiguity in my last change: "CPython's md5" takes...

We now act like CPython as of r6517. You'll need to convert non ascii 
(defaultencoding) input to raw strs first, like Java, or a 
UnicodeEncodeError will be raised