Vectorized pruning algorithm

[PhyloStar]

Added a PR. Please look at the functions. There is difference between the likelihood numbers from the example and the non-vectorized version.

[LinguList]

I modified the algorithm and the tests run perfectly (but beware to use the PyloTree, etc., see tests/test_pruning.py where I added the tests.

The basic idea is that we now have a unified call structure: algorithm(tree, patterns, matrix, characters).

We also have 100% test coverage now.

[LinguList]

If we add a Patterns class to pylodata (btw. you have to update to the most recent version of pylodata), the operation that is now in get_node_dict can also be done by the Patterns object:

def get_node_dict(tree, pattern, characters, transform=list):
    """
    Create a dictionary for the nodes which stores all states for all characters in a list.
    """

    return {leaf: transform([0 if char not in pattern[leaf] else 1 for char in
        characters]) for leaf in tree.root.get_leaf_names()}

This transformation is now moved to the scores dictionary in the main function, as it is anyway called, and we had redundant code there.

[LinguList]

BTW: using the dictionary to store pre-computed matrix lengths is also a rather enormous speed up of some 10 seconds.