Fix inconsistency in configuring VNET clusters

[thomasjungblut]

Right now the API exposes the following constructors for the client:

• HBaseClient(ClusterCredentials credentials)
• HBaseClient(int numRegionServers)
• HBaseClient(ClusterCredentials credentials, IRetryPolicyFactory retryPolicyFactory, ILoadBalancer loadBalancer = null)

The first call is for gateway deployments (agnostic of OS), the second by default for vnet load balancing on Windows clusters.

For linux clusters and VNET you would need to pass a load balancer in the third constructor with the right hostnames created. Despite not beeing able to instantiate a retry factory (see #43), this is quite inconsistent (why are the credentials needed?) and not nice for a user to setup.

Let's add a builder that creates a new HBaseClient:

var client = HBaseClientBuilder
   .New()
   .OnLinux()
   .OnVNet(numNodes: 32, loadBalancer: new LoadBalancerRoundRobin())
   .WithRetryPolicy(new ExponentialRetryPolicy(...))
   .Create();

Which will configure 32 nodes on linux with a vnet and RR balancing.

For the gateway case:

var client = HBaseClientBuilder
   .New()
   .OnWindows()
   .WithClusterUri(new Uri(...))
   .WithCredentials(new ClusterCredentials(...))
   .WithRetryPolicy(new ExponentialRetryPolicy(...))
   .Create();

We also need to refactor all the stuff in between, it has become a big mess with url strings getting passed around and no real structure.