MNT_WRITE_HOLD shouldn't leak into new vfsmount and neither
should MNT_SHARED (the latter will be set properly, along with
the rest of shared-subtree data structures)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>