datastore: Set namespace while initing a Dataset

[rakyll]

According to #85, namespaces are used to differentiate customer data. They could be seen as workspaces of the same data source. From that point, most of our users are likely to need namespaces most of the time.

What we currently do is to provide two different interfaces and key representations for namespace users. Rather than keep doing that, why don't we let them a namespace at the Dataset constructor.

var ds = new Dataset({
  projectId: '..',
  // ...
  namespace: '...'
});

Any operation can include a partition ID if a namespace is set.

[rakyll]

Our key representation will be an arbitrary list of <kind, identifier> tuples without namespace magic.