Use UTF-8 for Linux

[zeule]

The wchar_t in Linux is very wide, 32 bits, and thus almost unused, which, in turn, results in poor support in tooling. And that's to be expected when UTF-8 is the default. At the same time, Iconv, which is part of glibc, is universally available. Thus we can use it to convert UCS-2 into UTF-8 in Linux and always use UTF-8 in the library API.