Improved replica read command

Allow to specify strategy of selecting replica nodes. Now three strategies available:

• `:first` - synonym to `true`, previous behaviour and currently
  default. It means that the library will sequentially iterate
  over all replicas and will return as soon as find successful
  response, or report error from the latest.

c.get("foo", :replica => true)
c.get("foo", :replica => :first)
#=> "bar"
c.get("foo", :replica => :first, :extended => true)
#=> ["bar", 0, 11218368683493556224]

• `:all` - query all replicas in parallel. In this case the method
  will return array of the values on the all replica nodes without
  particular order. Also if the key isn't on the node, it will be
  skipped in the result array.

c.get("foo", :replica => :all)
#=> ["bar", "bar", "bar"]
c.get("foo", :replica => :all, :extended => true)
#=> [["bar", 0, 11218368683493556224],

1. ["bar", 0, 11218368683493556224],
2. ["bar", 0, 11218368683493556224]]

• `Fixnum` - you can also select specific replica node by its
  index in the cluster configuration. It should be in interval
  `0...c.num_replicas`

0...c.num_replicas
#=> 0...3
c.get("foo", :replica => 1)
#=> "bar"
c.get("foo", :replica => 42)
#=> ArgumentError: replica index should be in interval 0...3