Mdfreader

Adds .mfxz reading capability (zip compressed .mfx=mdf4 file)
Improved composition implementation, especially for array

several bug fixes but mainly introduces possibility to write mdf4 files.
By using mdf.write(), writes a file in the same major mdf version (3.x->3.3, 4.x-> 4.1) of the original.
If no file name given, original file name with '_new' appended is used.

New feature :
Warn user of unfinalised mdf4 file (from 4.11)

bug fix:
xlsx export robustness improvement regarding various kind of data
unsorted VLSD CG handling
convertToPandas bug fix for unsorted data

Major change : introduced mdf_skeleton class in mdf module in order to give flexibility of changing data structure without major impact on other modules. File metadata (date, time, author, etc.) are now accessible from file_metadata attribute. Also improved mdf representation with more concise print.
Improved memory consumption from both mdf3 and mdf4 by removing single file multiprocessing (mdfconverter still using file by file multiprocessing).
bugs fixes:
- thanks to mdf_skeleton, properly tackle unique name concern by appending datagroup number to channel name
- generally improved robustness of export methods

Corrected bug for non existing units in mdf4
Added dataRead cython extension to speed up by x60 processing of non byte aligned data (could not use numpy, used bitarray module at first but too slow). As cython requires compilation, I can not test all platforms, do not hesitate to feed back in case of issues.

Packaging for pip has been improved in this version. After install, mdfreader can be called from ipython and mdfconverter from prompt can launched.