Fix source annotation for file names with non-ASCII chars

For non-ASCII characters in source file names, KCachegrind
always assumed latin1 encoding. It is far better to assume
file name encoding taken from the systems locale (fromLocal8Bit).
In strange cases, this still may go wrong, but then the developer
also has other issues, and can convert the encoding of the
profile dump himself.

Fix suggested in bug 319143 by Vadim Ushakov.