This is the severalth time I've seen an attempt to hook up win32 Unix sockets to host Unix sockets. Which I think is reasonable in general, but I'm also not exactly thrilled about accepting this code without any tests whatsoever. Unfortunately, last time I tried to write tests for Unix sockets, I started triggering kernel panics in Windows. Maybe the situation has improved since then?
Of particular note is the difficulty in handling `sun_path`. Most of the functions that allow for translating Windows paths to Unix paths are not accessible from ws2_32.
I'm not sure I understand. wine_nt_to_unix_file_name() should be accessible from either side.
Although it's not quite clear to me that ws2_32 is the right place to do that translation; we may want to do it in ntdll instead.