we don't want open-by-handle to pick an in-core inode that
has failed setup halfway through.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>