Unify/simplify the MultiByteToWideChar() code and add wrappers for open()/stat()

[bit2shift]

This is the patch I mentioned on #1066 (comment).
Patches 2 subtle buffer overruns in:

// cchWideChar is the size, in characters, of the buffer indicated by lpWideCharStr.
// It was erroneously being passed twice that origina value, the corresponding number of bytes.
MultiByteToWideChar(CP_UTF8, 0, fname, -1, wfname, sz);
                                                   ^^
MultiByteToWideChar(CP_UTF8, 0, mode, -1, wmode, sz);
                                                 ^^

by way of unifying duplicate code into a separate function: utf8_to_utf16().
Also added a corresponding wrapper for open() -> _wopen().

Pending: dealing with the unsolved problem with modules located in Unicode paths.
Edit: dealt with.

[coveralls]

Coverage decreased (-0.02%) to 85.047% when pulling 13df637 on bit2shift:windows-unicode into ad9fc9f on stedolan:master.

[bit2shift]

Noteworthy: I switched from alloca() to jv_mem_calloc() since UNC paths have no length restriction.
Besides, alloca()'d memory in utf8_to_utf16() wouldn't be accessible to parent functions.