We already do error checking in nm_checkpoint_manager_create(). No need
to split it in two places. Let all error conditions be handled by
nm_checkpoint_manager_create() first, and then once we decide all is
good, nm_checkpoint_new() can no longer fail.