the original code was converting one character at the time from
whatever codepage of the input stream to unicode just to be able
to substitute the 'euro' character that apparently is somehow
not converted properly (why this is hacked here rather than fixed
in the code page converted is another story)

the change consist in translating the 'native' euro character in whatever
the converter think it should be, then converting the whole string at once
to Unicode and finally replace the 'bad' euro character representation
with the the 'good' one.

This turn the allocation pattern for O(N) to O(1) where N is the length
of the string.

Change-Id: Ie7ffeb813346953f539cf6dc34b43f890d8b970f