Create client library to reconnect / failover when using a cluster

[coryan]

Gee-H does not support multiple etcd servers, mostly because gRPC does not support them directly:

https://github.com/grpc/proposal/blob/master/A6-client-retries.md

we need to implement a client that wraps the gRPC calls and:

1. Maintains the current list of etcd cluster members, as that list can change over time.
2. Tries all cluster members before giving up on any operation.
3. Defines some mechanism to timeout to avoid too many retries.
4. Implements an exponential backoff strategy with jitter to avoid thundering herd problems when etcd servers restart.