Faster hDAG construction

[willdumm]

The method used to build the history DAG is slow (search entire DAG for duplicate nodes each time a history is added)

It certainly doesn't have to be this way. Let's think about this with the __getstate__ and __setstate__ implementations as inspiration.

[willdumm]

Certainly all DAG construction functions should be modified to expect only a generator containing trees, not an entire list, and iterate on that generator in a memory-efficient way.

Current situation:

Goal: build each node only once.
Input: list of ete trees?

• build node dictionary node_dict node -> node for existing DAG
• for each ete tree:
  • Postorder traverse to build child clades at all nodes
  • Preorder traverse, adding each edge by looking up parent node in the node_dict (it must be there by preorder), looking up child node in node_dict, adding if necessary, and adding edge from parent to child