improve performance of dtw

[maxbachmann]

These changes add some performance improvements to the dtw calculation. The main reason for the changes is the preparation for the transition to Cython. This way the actual transition requires less changes (e.g. the matrix access [a,b] can be optimized into direct accesses to the underlying matrix by Cython, which does not work when using the syntax [a][b])

@synesthesiam could you take a quick look at this?

[maxbachmann]

Using the examples provided in the tests these changes lead to the following performance improvements: _compute_optimal_path_with_window: ~20% _compute_optimal_path: ~40% compute_path: ~250%