Change how defaults are handled in the cluster:

 * master instance handles all defaults for all nodes
   * it is the one place with all defaults
   * depending on parameter style default is global for the cluster or can be
     specified per node with -frontend-config-<i>-parameter
 * kedifa and frontend nodes need everything to be passed, in order to
   simplify the request
 * configuration. defaults are dropped, as their scope is too large - they are
   not only defaulting the value on the master, but on each requested node
   which leads to extreme complexity of parameter passing
 * use extra_slave_instance_list only, as instance-slave-list is reserved
   parameter provided by slapos.cookbook:slapconfiguration, sent by SlapOS
   Master as slave_instance_list, to the master partition; enforce
   transmissions of slave list in extra_slave_instance_list and consider
   it's always present
 * useless replication_number return value is dropped