std.windows: use atomic rename, if possible

[matu3ba]

Mitigates #14978.

[squeek502]

This should use the same strategy as #16499, since the same caveat (presumably) applies for non-NTFS filesystems.

[matu3ba]

CI passes now. Do you think further tests are needed @squeek502 ?

[squeek502]

Would be good to test to confirm that this will actually mitigate #14978 (see the reproduction in #14978 (comment))

[squeek502]

I can confirm that the fallback works--the fs tests pass when running on a FAT32 drive (and I confirmed that INVALID_PARAMETER is being hit, so it's doing the proper fallback).

[squeek502]

I can also confirm that this successfully avoids AccessDenied for the reproduction in #14978 (comment)

> 14978-master.exe
error: AccessDenied
C:\Users\Ryan\Programming\Zig\zig\lib\std\os.zig:2664:27: 0x7ff764ae6e6a in renameatW (14978-master.exe.obj)
.ACCESS_DENIED => return error.AccessDenied,
^
C:\Users\Ryan\Programming\Zig\zig\lib\std\os.zig:2531:9: 0x7ff764ab76c7 in renameat (14978-master.exe.obj)
return renameatW(old_dir_fd, old_path_w.span(), new_dir_fd, new_path_w.span(), windows.TRUE);
^
C:\Users\Ryan\Programming\Zig\zig\lib\std\fs.zig:232:9: 0x7ff764ab1927 in finish (14978-master.exe.obj)
try os.renameat(self.dir.fd, self.tmp_path_buf[0..], self.dir.fd, self.dest_basename);
^
C:\Users\Ryan\Programming\Zig\tmp\14978.zig:5:5: 0x7ff764ab1814 in main (14978-master.exe.obj)
try af.finish();
^

> 14978-posix.exe

(there is a second process of the same .exe spawned at the same time for each of the above; the -master version hits AccessDenied every time, the -posix version [this branch] never hits AccessDenied)

---

Once #16717 (comment) is addressed, I think this will be good to go.