add ability to use a custom discovery strategy

Currently the only way to discover a hazelcast cluster is to either provide a static list of ip addresses or use cloud discovery. In an environment where the ip addresses of the cluster may change during the life of the python client, it would be great to specify a custom discovery strategy that is used to determine the member addresses dynamically at both initialization and reconnect events. This functionality would mirror the Java client which allows for the specification of a custom discovery strategy.