refactor _read_or_write

[bastibe]

This separates the data conversion for CFFI from the function selection.

This is a continuation of the discussion in #138 .

[mgeier]

I don't see an improvement from passing a string to passing a function object (while adding 2 new functions that only differ in the substring 'read' vs. 'write'), but I can live with it since it's only a minor implementation detail.

But the real problem is still unsolved: the error checking is separated from the code that relies on it.
I think this is solved in a less complicated manner in #138 (with less levels of indirection and IMHO easier to read), but if you prefer your way, you should at least call _check_array() inside of _read() resp. _write()!

[mgeier]

Probably we should skip the discussion here and implement #72 instead, because this potentially changes the same code (again) ...

But first, I think it would be a good time to release 0.7.1!

[bastibe]

superseded by #72.