tiny-count: efficiency improvements with strategic incorporation of Cython

[AlexTate]

There are several opportunities for improving the efficiency of tiny-count with Cython. However, Cython cannot be debugged using standard Python tools in its current stage of support/development (and the alternatives are troublesome for Mac users), so the application of Cython must be strategic to minimize the surface area for bugs. Candidates include:

• HTSeq's StepVector (this is already complete and testing shows a very significant improvement in runtime, but we will need to write our own ChromVector to wrap it due to issues with runtime patching for Python extensions)
• Classes in matching.py
• FeatureSelector.choose()
• FeatureCounter.assign_features()

Cython components will need to be moved into Cython-designated package files in order to minimize the extent of debugging preclusion

[AlexTate]

Closing this issue following the changes made in PR #218. Only the StepVector was Cythonized, but I think this will be the safest and most reasonable approach for the time being. Other targets for selection either didn't yield performance improvements sufficient to offset the maintenance costs, or were in components that are likely to undergo major changes in the near future.