Mdfreader

* Fixed bug swapping bytes order for signed int 8bytes with dataRead module.
* Time handling improvement: fileMetadata['time'] epoch floating in seconds is commonised for all mdf versions, no more dates as found in mdf version 3. More versatile this way and python datetime module can still be used to extract date if needed.
* Reworked merge_mdf to cope with heterogeneous data and renamed into concat_mdf
* Created merge_mdf

- Mainly dropping python2.7 support
- Switched from scipy to hdf5storage requirement for Matlab export with more recent format version 7.3
- Manipulation methods made compatible to mdf4
* Adjusted cut method to only cut data groups having same master type
* Rewrote resample in order to only resample data groups with same master type, added resample_group to resample only one specific data group.
- Fixed bug for attachment attribute that was empty for attachments and events in case of normal reading (noDataLoading=False)
- Improved parsing performance for VLSD and unsorted data by off loading heavy computations to cython module

Added possibility to write and read channels in column potentially for faster read.
Added half floating and complex types
Last release both compatible python 2.7 and 3.4+
Improved reading speed for SD Blocks (for variable length channels) and unsorted data

Fixed following bugs:
Several issues with VLSD_CG reading
bug when combining reading with no_data_loading and extracting data with raw_data, issue 168 , thanks bbto

fixed bug with dataRead having not aligned number of bytes.

Added parquet export
Added pandas dataframe export per data group
fixed a bug in dataRead with big endian channels

- /!\ Major refactoring for more pep8 compliance --> names of classes, methods and arguments have changed
- Added new method getChannelName4() for mdf4.x to only find mdf channel name (unique key within mdf class inherited from dict) based on name and source path or name (same name can be used in mdf file file but source paths or names shall be different, making it unique). Can be reused then with getChannelData4()
- Improved xml metadata parsing speed.

Fixed bugs related to
-nBytes calculation
-HL Block in VLSD data
-setup.py and cythonize
-channel by channel array reading implementation in python (fallback of dataRead)
-SD Block read of byte array type