Still a lots of details to change, but works.

Next is writing the feature and cost functions.

 * Gap costs within columns are yet to be defined.
 * Have a look at the timings !

multiple and structural purposes.

Gain of 40s (out of ~60) for PDB parsing time.

Faster than biopython, basis for multiple alignment.

It is not faster than biopython for the moment, but that's a start for
writing a decent multiple alignment function.