In-memory data store: NormalizedNode caching

We currently do not have an API which would allow us to reuse NormalizedNode trees in the data store. This is needed for users like BGP, which contain a large number of duplicate data (for example in attributes, AS PATH, etc).

Create a configurable DataStore/DataBroker decorator, which for specified (wildcard) InstanceIdentifier will maintain a Guava Cache and will transparently replace subtrees with shared versions.

Also devise a way how the BGP code can specify that an instance of this decorator should be placed between its BA DataBroker and the backing BI DataStore instance.