implement a progressive alignment integration algorithm

[GavinHuttley]

The limitation of dbga is that it requires a k that approaches the threshold of all k-mers within a sequence are unique, opposed by the requirement that the number of shared k-mers between sequences is maxmised.

The more sequences being aligned, the more this will (ultimately) increase the number of bubbles. My proposed "hack" is to:

• get a crude guide tree
• select triples through traversing the guide tree with each triple having a sequence in common with surrounding triples
• align triples of sequences
• merge triples with shared sequences using the align to reference capabilities of cogent3

[GavinHuttley]

This is a proof of principle implementation. It does improve alignment quality over a simpler align-to-reference approach in the small sample I ran it against, but I think that's due to some design decisions I made to simplif this first implementation (Knuth's dictum and all that).

Works off the Cogent3/tests/data/brca1.fasta and Cogent3/tests/data/murphy.tree files.

working.py.zip